 |
Description  |
|
|
TECHNICAL FIELD
This invention relates to heart rate monitors and, in particular, to a
heart rate monitor that is worn on a user's wrist and determines the
user's heart rate with medical grade accuracy during vigorous physical
exercise.
BACKGROUND OF THE INVENTION
Muscle contraction in the human body is caused by electrical biosignals.
Heart muscle contractions are caused by a biosignal referred to as the
electrocardiogram (ECG) signal. The QRS complex describes a region of
particular activity in the ECG signal during each heartbeat. Heretofore,
various devices have been designed to determine an individual's heart rate
by detecting the R-wave portion of the QRS complex.
These prior devices use essentially analog systems and may be described
with reference to FIG. 1. Referring to FIG. 1, a conventional heart rate
monitor 10 includes a sensor 12, that senses the QRS complex and
accompanying noise, and a filter 14 that produces a pulse each time the
QRS signal is detected. The pulse from filter 14 is amplified by amplifier
16 and compared with reference 18 by comparator 20. Whenever comparator 20
detects the proper condition, comparator 20 activates a monostable
multivibrator or "one-shot" 22. Digital logic circuitry 24 produces a
heart rate number to be displayed on LCD display 26 by counting the number
of times one-shot 22 is activated. The combination of filter 14, amplifier
16, reference 18, comparator 20 and one-shot 22 is a analog signal
processor.
Sensor 12, in the various prior devices, has been an electrode or a
microphonic, piezoelectric, photo-optical or capacitive apparatus. Filter
14 has included a filter or a series of filters. Reference 18 has been a
fixed voltage reference or inverse exponential voltage developed across a
discharging capacitor. Comparator 20 has been a standard comparator or
Schmitt trigger threshold circuit.
These primarily analog devices fail whenever the signal to noise (S/N)
ratio is low. The prior systems are inaccurate because in a noisy
environment they fail to detect certain signal pulses that represent heart
beats, and they falsely detect noise peaks as heart beats. The weaknesses
of prior devices are discussed in Leger, L., and Thivierge, M., "Heart
Rate Monitors: Validity, Stability, and Functionality", The Physician and
Sports Medicine, Vol. 16, No. 5, May 1988; and Allen, Douglas, "Heart-rate
monitors: The ideal exercise speedometer", Fitness Management, page 34-37,
Nov/Dec 1988. A low S/N ratio may be caused by at least three factors: (1)
a low amplitude biosignal of the user because of skin impedance which
varies from person to person when the person's skin is not properly
prepared, (2) vigorous exercise by the user, and (3) poor contact between
the sensor and the user.
Medical grade heart rate monitors, such as those found in hospitals, which
use chest electrodes mounted onto the user's chest, and an electrode gel
interface, do not ordinarily provide spurious results caused by low S/N
ratio problems. In these medical grade heart rate monitors, the signal
level is high because of the electrode gel and mounted electrodes, and the
noise level is low because the electrodes are mounted close to the user's
heart and, often, the user is not exercising while his heart rate is being
measured.
Wrist worn and hand held heart rate monitors, however, are susceptible to a
low S/N ratio because the inherently poor sensor contact with the user's
skin provides a low signal level. Moreover, the noise level is high
because the sensor is located remotely from the heart and the user is
typically exercising.
SUMMARY OF THE INVENTION
Accordingly, an object of the invention is to provide a heart rate monitor
that provides accurate heart rate measurement in noisy conditions where
the signal is of a relatively low level and the noise is of a relatively
high level.
Another object of the invention is to provide an accurate heart rate
monitor that may be used with no or minimal preparation of the user's
skin.
Yet another object of the invention is to provide an accurate heart rate
monitor that may be worn on the user's wrist or some other location remote
from the heart.
A further object of the invention is to provide a performance predictor
that indicates the speed and accuracy with which the heart rate monitor of
the present invention will work for the user so that the user may prepare
his skin or temporarily decrease his exercising to allow the heart rate
monitor of the present invention to perform quickly and accurately under
more vigorous exercise conditions.
The present invention relates to a heart rate monitor that determines a
user's heart rate with medical grade accuracy in noisy conditions of
exercise. The heart rate monitor includes sensors that sense a noisy
biosignal. The sensed signal is modified. Characteristics of a first
portion of the modified signal are learned. A second portion of the
modified signal is matched with the learned characteristics to produce a
matched signal. The user's heart rate is calculated from the matched
signal.
The invention determines the characteristics by correlating the first
portion of the modified signal with prestored biosignal data to produce a
correlated signal. A parametric screen and adaptive threshold are used to
determine whether the correlated signal represents a QRS complex. The
invention determines the user's heart rate by using the parametric screen
and adaptive threshold to determine whether a certain portion of the
matched signal represents a QRS complex. If it is determined that it does,
then an interval judge determines whether the time intervals between QRS
recognitions is consistent with what the user's heart rate should be. The
interval judge outputs a counter value that represents the average
interval time. A momentum machine converts counter value to rate
information. The momentum machine smooths and averages the rate
information and outputs it to a display.
A performance predictor determines how quickly and accurately the heart
rate monitor will work with a particular user by analyzing the signal to
noise ratio of the matched signal after it has been peak enhanced.
Additional objects and advantages of the present invention will be apparent
from the following detailed description of a preferred embodiment thereof,
which proceeds with reference to the accompanying drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a block diagram of a prior art heart rate monitor.
FIG. 2 is a graphical representation of a typical QRS complex region of an
ECG signal.
FIG. 3 is a block diagram of the hardware used in the present invention.
FIG. 4A is a graphical representation of actual data of a typical biosignal
before it is processed by the signal processing procedures of this
invention.
FIG. 4B is a graphical representation of actual data of a typical biosignal
after it has been processed by the signal processing procedures of this
invention.
FIG. 5 is an architectural block diagram showing in outline form the
procedures carried out in the digital pulse processor of the present
invention.
FIGS. 6A and 6B are detailed architectural block diagrams of the procedures
carried out in the digital pulse processor of the present invention.
FIG. 7 is an illustration of the measurement of certain parameters used in
the present invention.
FIG. 8 is an illustration of the processed signal and the threshold signal
of the adaptive threshold of the present invention.
FIG. 9 is a flowchart of procedures used in the interval judge of the
present invention.
DETAILED DESCRIPTION OF PREFERRED EMBODIMENT
Referring to FIG. 2, a typical QRS complex 50 is shown on biosignal 60.
Biosignal 60 is typical of an electrocardiogram ("ECG") signal detected
from a person's chest by using a standard 3-LEAD ECG cardiograph
instrument and chest electrodes mounted onto prepared skin. Skin is
prepared by rubbing an abrasive over the skin to remove dead skin and/or
by applying electrode gel to the skin. Q, R, and S designate sections of
the QRS complex. P and T designate particular activity in biosignal 60
before and after the QRS complex. The QRS complex varies from person to
person, but normal QRS complexes do not vary significantly from heart beat
to heart beat in the same person. As the heart beat of a person increases
or decreases, the time between QRS complexes decreases and increases,
respectively.
For ease of description, biosignal 60 will represent both a typical
noise-free biosignal and the noise-free biosignal of a hypothetical user
described in connection with FIGS. 3, 5, 6A, and 6B. Likewise, the signals
shown in FIGS. 4A and 4B represent both a typical preprocessed and
processed biosignal, and a preprocessed and processed biosignal of a
hypothetical user described in connection with FIGS. 3, 5, 6A, and 6B,
with signal pickup at the wrist and fingers without skin preparation.
HARDWARE
Referring to FIG. 3, a preferred embodiment of wrist worn heart rate
monitor 100 includes hardware apparatus 102. Hardware apparatus 102
includes electrode sensors 104A and 104B. Sensors 104A and 104B convert
biosignal 60 mixed with noise into a corresponding electric signal by
converting ions at the user's wrist and finger to electronic charge in
sensors 104A and 104B. Sensors 104A and 104B are electrically connected to
the differential inputs of instrument amplifier 110.
Heart rate monitor 100 is designed to be worn as a wrist watch with sensor
104B in contact with the user's wrist. To operate heart rate monitor 100,
a finger from the opposite hand is placed in contact with sensor 104A. For
example, if the left wrist is used, a finger from the right hand is used.
Heart rate monitor 100 is not limited to being used with a wrist and
finger. It may be used with any two parts of the body having opposite
polarity. For example, three other combinations that may be used are the
forehead and a finger on the left hand, both feet, and fingers from the
right and left hands. Therefore, whenever this application refers to the
user's wrist and finger, it is apparent that other parts of the body could
be substituted.
Further, while sensors 104A and 104B are electrodes in the preferred
embodiment, they may be any other type of sensor that detects a biosignal,
for example, a microphonic, piezoelectric, photo-optical or capacitive
sensor. Such sensors may be of the type described in United States Patent
No. 4,295,472 of Adams (electrode); U.S. Pat. No. 4,181,134 of Mason et
al. (electrode); U.S. Pat. No. 4,436,096 of Dyck et al. (microphonic);
U.S. Pat. No. 4,549,551 of Dyck et al. (microphonic); U.S. Pat. No.
4,195,642 of Price et al. (piezoelectric); U.S. Pat. No. 4,202,350 of
Walton (piezoelectric); U.S. Pat. No. 4,224,948 of Cramer et al.
(photo-optical); U.S. Pat. No. 4,301,808 of Taus (photooptical); and U.S.
Pat. No. 4,248,244 of Charnitski (capacitive). A silver chloride electrode
is preferred.
The noise mixed with biosignal 60 includes electrical noise sources, such
as common 60 Hertz, tribo electrical noise from cable flexing; machine
generated noise such as electromagnetic noise from exercising equipment;
and artifact noise from mioelectric muscle artifact, and surface artifact
at the electro-skin contact.
The corresponding signal from sensors 104A and 104B is amplified by
instrument amplifier 110 and filtered by filter 112. Filter 112 is an
analog band pass filter centered around 12 Hz with a band width of 6 Hz.
Filter 112 limits the frequency components of biosignal 60 mixed with
noise so as to satisfy the Nyquist sampling criterion when these signals
are digitized. Filter 112 also band limits low frequency noise. The
filtered corresponding signal is amplified by amplifier 114 and digitized
by analog to digital converter (A/D) 116. The output from A/D 116 is
preprocessed signal 120 shown in FIG. 4A.
Preprocessed signal 120 includes three QRS complexes associated with three
heartbeats, and noise. Preprocessed signal 120 is representative of the
signals that conventional heart rate monitors can produce without the use
of gels in making skin contact during exercise. FIG. 4A shows a
representation of actual test data of preprocessed signal 120 of heart
rate monitor 100.
Preprocessed signal 120 is processed by digital pulse processor 136, in
order to determine the user's heart rate. Digital pulse processor 136
includes a microprocessor that is controlled by system controller 140,
which also includes a microprocessor. The microprocessor in digital pulse
processor 136 may be an Intel 80C96 or 80C51. The software language used
to program the Intel 80C96 and 80C51 may be PLM, which is similar to PL1.
The microprocessor used in system controller 140 may be an NEC 7503. The
NEC 7503 may be programmed in machine specific assembly language.
The value of the heart pulse rate is displayed on LCD display 142. LCD
display 142 also includes a flashing heart shaped indicator that flashes
at the rate of the heart rate. In addition, LCD display 142 includes
information from a performance predictor that will be described in
connection with FIGS. 5 and 6B.
Ground element 106 is electrically connected to body common driver 150.
Body common driver 150 is an actively driven ground reference that is
isolated to protect the user against electrical shock. Ground element 106
is in close proximity to sensor 104A, such that the finger used to touch
sensor 104A will also touch ground 106. Whenever heart rate monitor 100 is
used in a wrist-worn configuration, it is preferable to not have a ground
in close proximity to sensor 104B because perspiration from the user's
wrist would tend to short out sensor 104B. In other embodiments, such as
an exercise bike handle bar-mounted sensor, a ground element is preferably
used in connection with both sensors 104A and 104B, since the ground and
sensor may be spaced sufficiently far apart to avoid shorting the sensors.
When a photo-optical system is used, sensor 104A would be a light source
and sensor 104B would be a light detector, with the user's skin placed
between the light source and light detector. If a photooptical system is
used, ground 106 is not needed. When a microphonic or piezoelectric system
is used, no electrical ground 106 is required. When a capacitive detection
system is used, sensor 104A could be a dielectric thin films coated metal
plate and sensor 104B could be ground. A separate ground 106 may be
needed. In the preferred embodiment, electrode sensors 104A and 104B
differentially pick up two different aspects of the same biosignal.
Alternatively, if a microphonic, piezoelectric, photooptical or capacitive
sensor is used, then there would not be a differential input unless more
than one sensor of a type is used, e.g., an array a photooptical sensors.
Sensors 104A and 104B are electrically connected to touch sensor 154. Touch
sensor 154 is an analog signal processor with a comparator that outputs a
logic one state to system controller 140 whenever both the user's finger
is in contact with sensor 104A and wrist is in contact with sensor 104B.
Touch sensor 154 outputs a logic zero otherwise.
User input 160 allows the user to select the mode of operation of heart
rate monitor 100 through system controller 140. Available modes include
time of day, stop watch and pulse mode. Pulse mode initiates a pulse read
function that determines the heart rate. Pulse mode is divided into three
modes: learning mode, operating mode, and intermediate mode. During
learning mode, the characteristics of the user's QRS complex are learned.
During operating mode, the learned characteristics are used in determining
the user's heart rate. During intermediate mode, the learned
characteristics are stored, but not utilized.
Learning mode is initiated when sensor 154 is in the logic one state
following the user's selection of pulse mode. Learning mode is concluded
whenever the characteristics have been learned, or either pulse mode
ceases to be selected or touch sensor 154 returns to the logic zero state
before the characteristics have been learned. Operating mode occurs
whenever the characteristics have been learned and retained in storage,
pulse mode is still selected, and touch sensor 154 is in the logic one
state. If the user continuously touches sensors 104A and 104B at the end
of learning mode, then heart rate monitor 100 will go straight from
learning mode to operating mode. Intermediate mode occurs whenever the
characteristics have been learned and retained in storage, pulse mode is
still selected, and touch sensor 154 is in the logic zero state. It is
contemplated that the user will switch back and forth between intermediate
mode and operating mode.
PROCEDURAL OVERVIEW
FIG. 5 is an architectural block diagram representation that provides an
overview of the procedures performed by digital pulse processor 136. When
the user wishes to begin determining his heart rate, he wears heart rate
monitor 100 so that his wrist is touching sensor 104B, selects pulse mode
on user input 160, and touches sensor 104A and ground 106 with his finger.
System controller 140 then initiates the pulse read function.
With digital pulse processor 136 in learning mode, math box 180 receives
preprocessed signal 120 from A/D 116. Math box 180 attempts to maximize
the S/N ratio of preprocessed signal 120 and enhances signal peaks through
a nonlinear transform. Math box 180 is a series of digital signal
processing algorithms. The algorithms performed by math box 180 may be
implemented in digital pulse processor 136 or in separate hardware.
Math box 180 determines the characteristics of the user's individual QRS
complex 50 by correlating preprocessed signal 120 with a standard QRS
complex representation that is stored in fixed reference 184. The output
from math box 180 is received through connector 186 by decision rules box
194. Decision rules box 194 is a series of decision trees that determine
whether the output of math box 180 contains a segment that represents a
QRS complex. If it is determined that a segment does represent a QRS
complex, then that segment is stored in learned user reference memory 206.
Once five segments representing the user's QRS complexes are stored in
learned user reference 206, an average of the five segments is made. A
template of the average is made and retained in learned reference 206. The
template represents the user's QRS complex. Digital pulse processor 136
then switches from learning mode to operating mode (although digital pulse
processor 136 may have been in intermediate mode and thus have gone from
intermediate mode to operating mode).
In operating mode, digital pulse processor 136 determines the rate of QRS
complexes from preprocessed signal 120. Math box 180 matches preprocessed
signal 120 with the user's QRS characteristics, i.e., the average QRS
complex template in learned reference 206, to maximize the S/N ratio of
preprocessed signal 120. The output of math box 180 on connector 186 is
processed signal 220, shown in FIG. 4B. Processed signal 220 is
substantially cleaner than preprocessed signal 120.
Processed signal 220 is received by decision rules box 194. Decision rules
box 194 determines whether particular segments of processed signal 220
represent a heart beat. Decision rules box 194 compares processed signal
220 with stored information in stored QRS and heart rate characteristics
registers 198 and information related to the user's current heart rate in
learned history register 226.
The output from decision rules box 194 is received by momentum machine 202.
Momentum machine 202 calculates the variable depth or weighted moving
average ECG frequency based on past and present ECG intervals. Special
techniques which are described in the procedural details section of this
application are used to smooth the data for a steady display. If the heart
rate is very high, then it is important to detect rapid changes in the
heart rate, and momentum machine 202 will respond accordingly by using
fewer samples in its average. If the rate is slow, then the user probably
does not demand quick response time. In this case, momentum machine 202
gives a slower, smoother response by using more samples in its average.
Communications channel 250 receives the result from momentum machine 202
and outputs it to LCD display 142 through system controller 140.
The ability of heart rate monitor 100 to determine an accurate heart beat
rate depends on three factors. The first factor is the signal to noise
ratio of the biosignal at the skin of the user. This ratio varies from
person to person. The second factor is the vigorousness with which the
user is exercising. More vigorous exercising causes greater noise and thus
decreases the signal to noise ratio. The third factor is the quality of
the contact between the skin of the user and sensors 104A and 104B. A
higher quality contact reduces the skin impedance thus allowing a greater
signal to be sensed by sensors 104A and 104B.
If one of the three factors is particularly low, heart rate monitor 100 may
still work if the other two factors are high. For example, if user's S/N
ratio is low, then the user could decrease the level of exercise or
increase the quality of the skin-sensor contact. The quality of the
skin-sensor contact may be increased by preparing the skin with an
abrasive or using electrode gel.
Heart rate monitor 100 includes a performance predictor 260. Performance
predictor 260 operates only during learning mode. Performance predictor
260 provides a performance number between 0 and 7 that is displayed on LCD
display 142. The performance number indicates the signal quality of the
user's processed signal 220. The performance number predicts how well
heart rate monitor 100 will perform for the particular user under the
noisy condition of exercise, i.e., how quickly heart rate monitor 100 can
determine the user's heart rate during typical exercise conditions. The
lower the performance number, the more quickly heart rate monitor 100 can
determine the user's heart rate during a given exercise activity, or
alternatively what amount of exercise activity can be tolerated while
maintaining reasonable read times. In addition, with a high performance
number, heart rate monitor may require more samples to determine the heart
rate in operating mode.
Heart rate monitor 100 provides medical instrument grade accuracy through
applying sensor electrodes 104A and 104B to the user's wrist or fingers.
Heretofore, medical instrument grade accuracy was only achievable through
the use of medical equipment including chest leads, electrode gel or
paste, and medical electrodes. Medical instrument grade accuracy is
defined in the Association for the Advancement of Medical Instrumentation
AAMI Standards and Recommended Practices 1987, Reference Book, paragraph
3.2.7, section EC13:
3.2.7 Range and Accuracy of Heart Rate Meter. The minimum allowable heart
rate meter range shall be 30 to 200 bpm, with an allowable readout error
of no greater than .+-.10 percent of the input rate or .+-.5 bpm,
whichever is greater. Cardiac monitors labeled for use with pediatric
patients shall have an extended heart rate range of at least 250 bpm. In
addition, input ECG signals at rates less than the disclosed lower limit
of the rate meter range shall not cause the meter to indicate a rate
greater than this lower limit. Input signals at rates above the disclosed
upper limit of the rate meter range, up to 300 bpm (350 bpm for monitors
labeled for use with pediatric patients), shall not cause the meter to
indicate a rate lower than this upper limit.
PROCEDURAL DETAILS
FIGS. 6A and 6B show a more detailed representation of the procedures that
are included in FIG. 5. FIG. 6A shows the procedures of FIG. 5 that occur
before processed signal 220 is received by decision rules box 194. FIG. 6B
shows the procedures of FIG. 5 that occur after processed signal 220 has
been received by decision rules box 194. Referring to FIG. 6A,
preprocessed signal 120 is received by digital filter 264 from A/D 116.
Digital filter 264 is a band pass filter centered at 12 Hz with a band of
about 6 Hz. Digital filter 264 may be recursive or nonrecursive.
Referring to learning mode, when heart rate monitor 100 is first used,
switches 268A and 268B are connected to cross-correlator 272.
Cross-correlator 272 is a fixed reference correlator implemented using
integer arithmetic, although floating point arithmetic may be
alternatively used. Fixed reference 184, as explained in connection with
FIG. 5, is a numerical representation of a standard QRS complex. Fixed
reference 184 may have from 16 to 32 data points, depending on the data
sampling rate.
Cross-correlator 272 receives an incoming bit stream from digital filter
264 and correlates the entire bit stream as each new bit is received. The
output of cross-correlator 272 is a maximum peak whenever the incoming bit
stream correlates closest with the contents of fixed reference 184.
Cross-correlator 272 takes the sum of the bits multiplied by a
corresponding value from fixed reference 184 as described by equation 1
below:
##EQU1##
where y is the output of cross-correlator 272, C is one of n coefficients
from fixed reference 184, and x is a particular sample of the signal from
digital filter 264. The parameter d represents the number of datapoints
and is between 16 and 32.
Peak enhancer 276 performs a nonlinear transform on the signal from
cross-correlator 272 according to equation 2 below:
f(x)=.vertline.x.vertline..sup.n sign(x) (2)
where n is an empirically chosen constant of between 1.6 and 2.0. The
absolute value of x is raised to the power n. In order to retain the
original signal of x, .vertline.x.vertline..sup.n is multiplied by the
sign of x. The value of n is preferably chosen to be 2.0 for simplicity
and efficiency in real-time computing. However, other values may be more
optimal where the signal peaks received by peak enhancer 276 vary broadly
in amplitude. Alternatively, the constant n could be an adaptive variable
that is set under control of performance predictor 260 to allow
optimization if the sample rate is too low or if QRS complex 50 is
strongly modulated by a respiration wave.
The purpose of peak enhancer 276 is to further increase the numerical value
of the peaks of the signal from cross-correlator 272, thus accentuating
the difference between the signal peaks and the noise. The output of peak
enhancer 276 is processed signal 220, shown in FIG. 4B. Symbol A connects
FIG. 6A to FIG. 6B.
An automatic gain control AGC 278 may be used with heart rate monitor 100
when floating point arithmetic is used. AGC 278 may be used with an
autoscaler and overflow detection when integer arithmetic is used.
Adaptive threshold 288 performs an equivalent function to AGC 278. AGC 278
sets a gain range and adaptive threshold 288 works within that range.
Processed signal 220 is received by parametric screen 282. Parametric
screen 282 is activated and begins its measurement process when an
internal state machine determines that the incoming waveform data stream
has transitioned through a local minimum turning point followed by a local
maximum turning point, followed finally by a local minimum turning point.
Waveform alignment occurs at the last turning point and parametric screens
are applied. This alignment point will become the point of detection if
the waveform passes the parametric screen and meets requirements of
adaptive threshold 288. Parametric screen 282 is a decision tree that
determines whether processed signal 220 has slopes, widths, and amplitudes
that are within standard ranges as will be explained in connection with
FIG. 7.
FIG. 7 shows a portion of processed signal 220 that represents a QRS
complex that has been received by parametric screen 282. Parametric screen
measures processed signal 220 at certain points to determine D.sub.1,
D.sub.2, M.sub.1, M.sub.2, and M.sub.3 as are shown in FIG. 7. If any of
these parameters is not within the prestored standard range, then a signal
is not output from parametric screen 282. Alternatively, if one or more of
these parameters narrowly miss the range, but other parameters are within
the range, parametric screen 282 could allow processed signal 220 to pass.
If processed signal 220 does not meet the requirements of parametric screen
282, then it does not advance further in heart rate monitor 100. If
processed signal 220 meets the requirements of parametric screen 282, then
it is subject to the requirements of adaptive threshold 288.
Adaptive threshold 288 contributes to the accuracy of heart rate monitor
100 by reducing the number of false positive or false negative detects
that occur. A false positive detect occurs when a heart rate monitor
determines that a portion of a signal represents an R-wave of a QRS
complex when in fact it does not. A false negative detect occurs when the
detector fails to fire when a valid R-wave is present. Prior art hardware
techniques use a fixed reference threshold or a negative exponential
decaying voltage on a capacitor. The software thresholding techniques used
by adaptive threshold 288 reduce errors by applying adaptive algorithms to
peaks in the signal from parametric screen 282. The algorithms used in
adaptive threshold 288 are explained in connection with FIG. 8 and
equation 3 below:
##EQU2##
where 0.6.ltoreq.k.ltoreq.0.75 depending on the average noise level,
.lambda. is the inverse of the time constant (e.g., 4 or 5 seconds)
divided by the data sample rate (e.g., 125 samples/second). T is the
number of samples since the last QRS detection, R-wave-peak is the
magnitude of the R wave in the QRS signal and n is the size of the array
that stores R-wave-peak values. A time constant of 4 seconds and sample
rate of 125 samples/second would result in .lambda.=0.002. However,
.lambda. may be 0.001955 for programming ease since the PL1program
implements e.sup.-.lambda.t as a shift operation.
FIG. 8 shows processed signal 220 and QRS complexes 290A-290E. Threshold
291 is shown from time t.sub.0 to time t.sub.4. At time t.sub.0, threshold
291 is initialized to its starting value. Threshold 291 falls according to
equation 3. If a detection has not occurred within two seconds, then
threshold 291 is set to k multiplied by the largest peak that has occurred
within the last 2 seconds. Time t.sub.1 occurs 2 seconds after time
t.sub.0. However, if a peak crosses threshold 291 during the first two
seconds, then threshold 291 is set to k multiplied by that peak.
Subsequently, each time a valid QRS complex is detected at times t.sub.2,
t.sub.3 and t.sub.4, threshold 291 is reset to k multiplied by the average
of last n peaks. Alternatively, the threshold may be set to the median
value of the last n peaks to provide a marginal improvement in the
| | |
