Designing a web-based health information system

Designing a Web-based Health Information
System Using Photoplethysmogram (PPG) Signal
Abstract—This paper researchs about the non-invasive methods of measuring health parameters based on the optical signal
Photoplethysmography (PPG). Then, this paper proposes the
methods of processing PPG signals using Wavelet transform.
From the theoretical research, this paper proposes to design
and build a web-based system to monitor the health parameters
using PPG signals. The PPG signals will be processed on
SAM4E-EK board to obtain the desired results, then they are
stored in a database and displayed on the website for remote
monitoring.
Index Terms—photoplethysmography, PPG, NIBP, SAM4EEK, health information, non-invasive, blood glucose, blood
pressure, heart rate.

I. I NTRODUCTION
With the development of science and technology, there are
plenty of personal health parameters monitor (e.g. heart rate,
blood pressure, oxygen saturation concentration, ...) which
have been produced and used widely. However, most of
users do not have much knowledge of medicine. Perhaps
they can know this parameters, but they cannot know their
health status definitely. And in the other hand, for example,
people with heart disease, although patients can understand
the state of their heart rate, but when one has a heart attack,
without person next to do some helps, he will be in a very
dangerous situation. So a problem arises that how we can
monitor patients’ health status remotely to be able to solve
the situation. This paper will research to build that system.
An important problem in the process of the system above
is the methods to determine the parameters of personal
health. The classical methods are often complicated, require
the assistance of doctors so it is difficult to use them
individually. For example, the method of determining blood
glucose. Normally to measure this parameter, patients must
take blood from the fingertip, this often causes pains and
risks. The classical method for measuring blood pressure
using pressure cuff manually depends on the experience
of doctors of nurses, therefore it is not used widely for
individuals.
Many methods have been proposed to solve these problems, such as Photoplethysmography, Electrocardiography,
Electromyography, .... In which, the photoplethysmography
- PPG method has been widely used. PPG [1] is an optical,
non-invasive measuring technique that provides immediate
diagnostic information on the cardiovascular state. PPG
signals reflect the change in blood volume caused by blood
vessel expansion and contraction, which can be detected by
a photodiode if external light is illuminated into the tissue.
Time intervals between the PPG peaks can be converted into
the heart rate or pulse, but the waveform characteristics of
each individual PPG pulse contain hemodynamic information. The PPG signal consists of two components - a slow,
varying DC offset representing the skin: blood volume in the
probe-covered area, and a fast, alternating AC component
that reflects the blood volume pulsations. The amplitude of

the AC-component is only 0,5-4% of the DC offset in a PPG
signal.
The advantage of the method measuring health parameters based on PPG signal which is simple in theory, low
implementation costs, can be done automatically and easy
deployed in practice widely. In this paper, the method of
determining the value such as: heart rate, blood pressure,
blood glucose-based PPG signal will be introduced. Based
on theoretical study, some methods will be implemented
in MATLAB simulation and on the hardware. The PPG
signals used for determining blood pressure in this paper
are measured on mice and provided by the laboratory
MABEL (Mannheim Biomedical Engineering Laboratories),
Mannheim University [2]. In addition, this laboratory also
provides us with SAM4E-EK board for the construction of
the system. This is a powerful board from Atmel. We will
introduce more about this board lately.
Method of determining the value of heart rate using PPG
signal is implemented on real hardware, the results are
checked with the actual device OMRON. For the method
of determining blood glucose based on PPG signals, this
method is relatively new and is still continuing research. At
present there are only a few actual models of this method,
but they all require complicated techniques. Therefore in this
paper, the method of determining the blood glucose just stop
at theoretical research.
One of the important aspects of research methods is
the PPG signal processing. The majority of signal analysis
methods currently use Fourier analysis. But in practice,
sometimes Fourier analysis has limitations, especially for
unstable signals (non-stationary) as PPG signal. Therefore
this paper is focused on the application of Wavelet transform
in signal processing. Continuous Wavelet Transform (CWT)
and Discrete Wavelet transform (DWT) will be studied to
determine the value of heart rate as well as remove noise
from PPG signals. The rest of this paper focuses on building
a web-based model to monitor personal health parameters
based PPG signal remotely . This process includes: designing
a simple device to measure health parameters; applying
wavelet-based denoising theory in the equipment; building
a Java-based software to display health information as well
as transmit data over the Internet; build a website to monitor
the health parameters remotely.
II. A N ON -I NVASIVE B LOOD P RESSURE M EASUREMENT
M ETHOD U SING P PG S IGNAL
A. Principles
A heterogeneous selection of mice (different age, weight
and genealogy) was taken into account in order to validate
the new method without requiring drugs for a specific
increase or reduction of BP. Figure 1 illustrates a schematic
diagram of the device.

Fig. 1. Schematic diagram for indirect measurement of mouse blood
pressure.

The system consists of two photodiodes a separated a
minimum 15 mm from each other. This distance is based on
the minimum detectable PTT of our system (3 ms) and the
maximal pulse wave velocity reported in mice (4.60 m/s) [3].
They record the reflected light of an infrared LED attached
to the tail of the mice. At the same time, a pressure cuff is
initially inflated to a pressure (180 mm Hg) that exceeds the
systolic arterial pressure of the mice. It is then reduced by
a rate of approximately 2.5 mm Hg/s. The recorded signals
of the photodiodes are then used to estimate the PTT curve,
this curve will be later correlated with the recorded signal
of the cuff pressure. The method described in this paper can
also be implemented in a transmission mode if the distance
between the light source and the sensors is smaller than 1
cm.
The result of the external pressure applied to the artery by
the cuff is the modulation of the artery characteristics such
as the compliance. This modulation can be described through
the formula equation (1) [4].
C(Pi , Pcuf f ) =

2
Rmax
(
P1

1
)
Pi − Pcuf f 2
1+(
)
P1

(1)

P1 : relates to the slope at the inflection point, Pi : internal
pressure, Pcuf f : cuff pressure, Rmax : is the maximal asymptotic vessel radius.
Based on the Bramwell-Hill equation (2) can be demonstrated the relation between the PTT, the compliance, and
cuff pressure [5].
s
d
1
v=
=
(2)
PTT
ρC(Pi , Pcuf f )
ρ: blood density, d: distance between sensors.
Figure 2 shows the relation between the PTT and compliance. The three BP values are indicated.
The method used in this study takes advantage of this
effect in order to estimate the BP indirectly. The PTT is
calculated between the recorded signals of the photodiodes.
It is obtained from the sensor nearest to the heart (proximal)
and the signal of the sensor situated further away from the
heart (distal). It shows the three characteristic features for
the determination of the systolic, the mean and the diastolic
BP (figure 3).
The following paragraph explains the method developed
on which we based this study.

Fig. 2. Modulation of the artery characteristics: compliance and PTT as
function of the internal and the cuff pressure. It shows too the three BP
values.

Fig. 3. Characteristic points in the PTT curve. The descending curve shows
the cuff pressure.

•

•

•

•

The cuff is initially inflated to a pressure of 180 mmHg
that exceeds the systolic arterial pressure of the mouse.
As a result of the external pressure, the artery is totally
compressed and therefore the blood cannot circulate and
thus the pulse cannot be detected. The cuff pressure is
reduced by a rate of approximately 2.5 mm Hg/s.
At the time at which the cuff pressure is the same as
the systolic BP, the blood can circulate again and the
wave appears distal of the cuff, the transit time is at its
minimum point. The increase of the transit time curve
is considered a characteristic feature.
Maximum of the transit time curve. At this point,
the cuff pressure is equal to the mean BP. Due to
this pressure; the compliance of the artery is maximal.
Hence, the pulse wave velocity is minimal while the
PTT is maximal [6].
Once the compliance reaches the maximal point it starts
to decrease and with it the PTT. When the cuff pressure
is the same as the diastolic value, the modulation of
the artery be the cuff is completed and the compliance
remains constant. The buckling in the PTT curve is the
diastolic BP.

B. Results
Figure 4 shows a typical change of the transit time
compared to a simultaneous direct BP measurement during
the deflation phase of the cuff. The calculation of the indirect
BP is based on the beat to beat measurement of the transit
time and the locations of the three characteristic features:
systolic, diastolic and mean blood pressures.
Figure 5 shows an example of the recordings of the proximal and distal signals (A), the PTT curve as a result of the

TABLE I
S TATISTICAL C OMPARISON B ETWEEN D IRECT AND I NDIRECT B P
M EASUREMENTS
DBP

No.

Fig. 4. Typical PTT alterations during a deflation phase of the cuff (Xaxis) compared to the direct measurements (right Y-axis). The vertical lines
show the BP determined by the interpolation with the cuff pressure. At the
100 mm Hg cuff pressure the blood could circulate and the transit time is
at its minimum point, it is the systolic pressure. At the 80 mm Hg cuff
pressure, the pressure is equal to the mean pressure and the PTT curve has
a maximum. At the 60 mm Hg cuff pressure the PTT curve has a buckling
and it is the diastolic pressure.

time difference between both signals with the corresponding
BPs (B) and the cuff pressure (C).

MBP

SBP

PTT

Direct

PTT

Direct

PTT

Direct

1

123.7

121

122.5

118

118.5

111

2

116.8

114

113.1

110

106.6

101

3

131.4

133

130.2

128

122.5

120

4

118.5

120

107.8

115

98.06

97

5

116.5

118

109.5

112

96.44

98

6

118.3

121

115.2

112

106.8

105

7

93.8

99

85.45

82

70.2

72

8

113.7

111

96.52

104

88.3

93

9

127.4

129

107.8

118

97.07

103

10

127.8

127

120.8

119

98.06

107

III. A N ON -I NVASIVE B LOOD G LUCOSE M EASUREMENT
M ETHOD U SING P PG S IGNAL
A. Theoretical Analysis
1) Molecular Composition and Relative Absorptivities:
When a photon is incident on a molecule, there will be
bond deformations or bond vibrations at different energy
levels related to different bonds, depending on the energy
of incident photon [7]. So, only the photon with energy that
corresponds to the difference between two of its energy levels
can be absorbed. The frequency of the vibration is given by
the equation (3).
r
1
k
ϑ=
(3)
2π m
where k is the bond strength and m is the reduced mass.
For a glucose molecule, the molecular structure is as
shown in figure 6.

Fig. 6. Molecular structure of D-glucose.

Table II shows the frequencies corresponding to different
bond vibrations in glucose molecule [8].
Fig. 5. An example of the recorded distal and proximal signals (A). PTT
curve with the SBP, MBP and DBP points (B). Cuff pressure curve with
the corresponding BP points (C).

Value differences can be explained by the fact that there
is a certain significant distance between the measuring positioning of the sensors and the heart. Therefore, the indirect
systolic BP values are higher and the mean and diastolic
BP values are lower than those of the direct measurements.
In order to be able to compare our measuring values to the
values described in the literature, the average BP values from
several measuring cycles of a session were calculated. With
this procedure, the comparison of the blood pressure-value
pairs is more related to the mice than to the measuring cycles.
Table I summarizes the comparison between the direct and
the indirect measurements.

TABLE II
A BSORPTION BANDS OF G LUCOSE B ONDS
Overtone

Wavelength

Bond

Wavelength

Bond

Fundamental

3377nm

vC-H

2817nm

vO-H

First overtone

1688nm

2vC-H

1408nm

2vO-H

Second overtone

1126nm

3vC-H

939nm

3vO-H

At a deeper level absorption of light can be seen as
dependent on the probability of absorbance of a photon by
the molecule. For nth overtone final energy is (n+1)E, where
E is the fundamental energy. As n increases, probability
of absorbance decrease rapidly and hence intensities of
absorbance decrease as overtones increase. The absorption
at fundamental frequency is calculated and from that the

absorption at second overtone is calculated relatively [9].
Table III depicts the relative absorptivities of C-H bond at
different overtones with respect to the fundamental. It is
evident that the absorptivity decreases rapidly as one goes
from fundamental to the overtones in sequence.
TABLE III
R ELATIVE A BSORPTIVITIES OF C-H B OND
Overtone

Wave number cm−1

Relative Absorptivity

Fundamental

3019

1

First overtone

5912

0.088

Second overtone

8677

3.2 × 10−3

Fig. 8. Graph showing the relative sensitivity of light emitted by 1070nm
LED.

2) Selection of the Wavelength of Peak absorption: The
absorption spectrum of the glucose has been studied in order
to choose the wavelengths for LEDs. For this purpose a IR
absorption spectrum of 0.1M aqueous glucose solution has
been collected and analyzed in second overtone region of
the near-infra red spectra using the Bruker tensor 27 FTIR
spectrometer. Figure 7 shows the absorption spectrum of the
glucose over the second overtone region where the selected
wavelengths are shown.

Fig. 9. Intensity weighted mean absorbances over 920nm to 1100nm.

range of LED is high. For the current research, wavelengths
of 935nm, 950nm, and 1070nm have been chosen.
B. Method
According to Beer-lambert’s law the absorbance of light
by a liquid is related to the concentration of the material by
equation (5).
A = Cl
(5)
Fig. 7. Spectrum showing the transmittance of glucose.

We can observe that the absorption peaks in this region
are very narrow typically of the order of the 2nm to 5nm but
the LED emits the light over a range of wavelengths. The
wavelengths are chosen such that the weighted average of the
absorption over the spectral bandwidth of the LED is high.
While calculating this weighted average the intensity of light
emitted by the LED acts as weight for the absorption at that
particular wavelength. The expected value of the absorbance
over a range has been calculated as equation (4).
Z ∞
E[A] =
A(λ) × f(λ) dλ
(4)
−∞

where A is the absorbance of glucose over a range of
wavelengths and f (λ) is the probability density function of
intensity of light emitted.
By multiplying the relative intensity of the light emitted
by the LED shown in figure 8 and the absorbance value
of glucose taken from the spectrum and averaging it over
spectral distribution of LED we calculate the mean value of
the absorbance which is shown in figure 9.
In order to improve the accuracy of the prediction of blood
glucose level, the wavelengths have been chosen such that
mean value of the absorbance of glucose over the frequency

where  the molar absorptivity of solute at a particular
wavelength, C is is the concentration of the solute and l is the
path length. Hence for a specific wavelength i, equation (5)
may be written as
Ai = i Ci li
(6)
In our particular case, i = 1 corresponds to a wavelength
of 935nm, i = 2 corresponds to a wavelength of 950nm
and i = 3 corresponds to a wavelength of 1070nm. As the
concentrations and path lengths are same for a person at a
particular time, we can write
C1 = C2 = C3

(7)

l1 = l2 = l3

(8)

As discussed in the previous section, three different wavelengths have been selected which have peak absorption of
glucose in the near infrared region of the spectrum, 1070, 950
and 935nm. For getting glucose level in blood the absorbance
which is mainly due to blood glucose is calculated from AC
component of PPG wave as follows [10]
∆ODλ = log[1 +

∆Iλ ti
]
Iλ ti+1

(9)

where ∆ODλ is the difference between optical densities at
time ti and ti+1 , ∆Iλ ti is the pulsatile component at time
ti and Iλ ti+1 is the intensity of light at time ti+1 .

This difference has the effect of removing the venous
and the tissue contributions to yield only the change in
intensity due to the pulsating arterial blood compartment
[11], [12]. The optical densities have been calculated at three
wavelengths for 35 subjects. The actual glucose values have
also been determined using standard invasive glucometer.
Regression analysis has been done on these values using
neural network toolbox in MATLAB.
C. Regression Analysis Using Artificial Neutral Networks
Multivariate calibration models have come into wide use
in quantitative spectroscopic analyses due to their ability to
overcome deviations from the Beer Lambert’s law caused by
effects such as overlapping spectral bands and interactions
between components [13]. PLS and PCR are the most widely
used chemo-metric techniques for quantitative analysis of
complex multicomponent mixtures. These methods are not
optimal when the relationship between the IR absorbance’s
and the constituent concentration deviates from linearity.
The theory and the application of Artificial Neural networks
(ANN) in modeling chemical data have been widely presented in the literature [14]. In this current work, ANN has
been used for function fitting to develop a model based on the
PPG data and invasive glucose measurements. For calculating
glucose concentration from beer-lambert’s law we can say
that
X
A[{Ci }, λ] =
(Ci Ai (λ))
(10)

noises and artifacts in order to support clinical decision
making. To address this issue, DWT allows effective noise
reduction. The DWT splits the signal into two components,
each of half the original length, with one containing the
low-frequency or “smooth” information and the other the
high-frequency or difference information. The process is
performed again on the smooth component, breaking it
up into “low-low” and “high-low” components and it is
repeated several times. The simpler way to remove noise or
to reconstruct the original signal from a contaminated signal,
in case of 1D or 2D, using the wavelet coefficients which
are the result of decomposition in wavelet transform, is to
eliminate the small coefficient associated to the noise. After
updating the coefficients by removing the small coefficients
assuming as noise, the original signal can be obtained by
the reconstruction algorithm using the noise free coefficients.
Because it is usually considered that the noise has high
frequency coefficients, the elimination of the small coefficient generally applied on the detail coefficients after the
decomposition. Indeed, the main idea of the wavelets denoising method to obtain the ideal components of the signal
from the noisy signal requires the estimation of the noise
level. The estimated noise level is used in order to threshold
the small coefficient assumed as noise. The noise reduction
procedure follows the flowchart in figure 10.

i

where Ai (λ) is the optical density at wavelength λ, of
the ith component, whose concentration is Ci , A(λ) the
optical density of the mixture. If we have the spectra of all
the components {Ai (λ)} and measure the spectrum of the
mixture {Ai (λ)}, we can use a mathematical tool to estimate
the {Ci }. The three wavelengths have been chosen such
that glucose has considerable amount of absorption when
compared to other components. So we have given the matrix


Aλ1i=1...25
Aλ2i=1...25 
(11)
Aλ3i=1...25


as input and Ci=1...25 as output which are measured using
standard invasive method. In this way we have three different
optical densities as inputs and one output for a single sample.
MATLAB neural network toolbox has been used. A two
layer feed forward network with sigmoid hidden neurons and
linear output neurons has been chosen. Number of hidden
layers have been chosen as 10.
IV. A PPLICATION OF WAVELETS IN P PG S IGNAL
P ROCESSING
A. PPG Signal Improvement Using DWT
The PPG signal profile has a very small amplitude, so
some artifact can arise due to patient motion during measurement that are of comparable amplitude, as well as “transport
artifacts” arising when a patient is being transported by
ambulance. The other artifact can be patient’s body vibration,
which can have large amplitude and a frequency close to the
patient heart rate. This close proximity in frequency makes
removal of vibration artifact from PPG signals especially
difficult by Fourier techniques. The PPG signal has been
used for recognizing much variability’s of health, so it is very
important to get the parameters of this signal clear without

Fig. 10. The flowchart of the noise reduction algorithm.

The crucial issue of this approach is determination of an
appropriate threshold value. In this paper, the authors applies
the following hard thresholding rule on the coefficients ci,k .
(
ci,k , |ci,k | ≥ λi
c˜i,k =
(12)
0,
|ci,k | ≤ λi
where λi is the threshold that depends on the noise level at
the ith level; the signal is then reconstructed by the IDWT
of the c˜i,k coefficients. It is essential to find the appropriate
value for the threshold. In this paper, we use the universal
threshold estimator proposed by Donoho. It uses a fixed
threshold form given as:
p
(13)
λ = σ 2 log n
where n denotes the length of the analyzed signal and σ is
given by:
median(dL,k )
σ=
(14)
0.6745
where dL,k are the coefficients at the highest decomposition
level.
In this section, the authors implement the noise reduction
algorithm in MATLAB. Firstly the considered signal was
decomposed using a five-level wavelets decomposition. The
reason for five-level decomposition is that it is the maximum
level that the approximation component is least distorted
compared to the original signal. The other factor deciding
this method’s efficiency is the mother wavelet. One of the

key criteria of a good mother wavelet is its ability to fully
reconstruct the signal from the wavelets decompositions. We
used in our analysis the Daubechies db4. Figure 11 shows
DWT results with all approximation and detail components.

scale and frequency. Obviously, there is clearly a relationship
between scale and frequency. Recall that higher scales correspond to the most ”stretched” wavelets. The more stretched
the wavelets, the longer the portion of the signal with which
it is being compared, and therefore the coarser the signal
features measured by the wavelet coefficients. To summarize,
the general correspondence between scale and frequency is:
•
•

Fig. 11. Signal decomposition using db4 mother wavelet.

The high frequency components of the signal decreases
as lower details are removed from the original signal. As
the lower details are removed, the signal becomes smoother
and the noises disappears since noises are marked by high
frequency components picked up along the ways of transmission. This is the contribution of the discrete wavelet
transform where noise filtration is performed implicitly.
Figure 12 presents the signal before and after denoising.

Low scale a ⇒ Compressed wavelets ⇒ Rapidly changing details ⇒ High frequency ω.
High scale a ⇒ Stretched wavelets ⇒ Slowly changing,
coarse features ⇒ Low frequency ω.

While there is a general relationship between scale and
frequency, no precise relationship exists. Users familiar with
Fourier analysis often want to define a mapping between a
wavelet at a given scale with a specified sampling period
to a frequency in hertz. You can only do this in a general
sense. Therefore, it is better to talk about the pseudo frequency corresponding to a scale. The Wavelet ToolboxTM
in MATLAB provides two functions centfrq and scal2frq,
which enable you to find these approximate scale-frequency
relationships for specified wavelets and scales. The algorithm
of this method follows the flow chart below (figure 13).

Fig. 13. The flowchart of the frequency estimation algorithm.

Fig. 12. Removing background noise from the PPG signal.

In addition, the PPG signal artifact removal algorithm
also makes use of DWT. This algorithm is based on the
observation that the discrete wavelet transform puts the
biomedical waveform in a very different region of the
transform plane than the signal components attributable to
artifact. This can be seen clearly by a visualization of the
discrete wavelet transform in which the absolute values of
the wavelet transform coefficients, as a function of time
and level of the transform (scale), usually called time-scale
plot, or scalogram. The artifacts, if show up, usually present
most prominent in the short scale (high level of transform)
portions of the scalogram. So in order to remove the artifacts,
these portions should be zeroed out. The rest of the DWT
coefficients may then be reconstructed to be used as an input
to an PPG pressure determination algorithm in the usual way
for the measurement of desired patient pressure values.
B. PPG Signal Frequency Estimation Using CWT
In order to estimate PPG signal frequency using CWT,
there is two problems that we have to clarify, they are

The important factors influencing algorithm’s result are
scale and Wavelet family. In our calculation, signal’s frequency is usually between 5 and 15 Hz, so we choose
scale with the minimum value is 50, the maximum value
is 256, and the step is 0.1. The other important condition affecting the result of wavelet transform is the base
(mother) wavelet.The advantages of wavelet transform for
signal analysis is the abundance of the base wavelets. From
such abundance arises a natural question of how to choose a
base wavelet that is best suited for analyzing a specific signal.
Since the choice in the first place may affect the result of
wavelet transform at the end, the question is valid.
The complex-valued Morlet’s wavelet is often selected
as the choice for signal analysis using the CWT. Morlet’s
wavelet insures that the time-scale representation can be
viewed as a time-frequency distribution. This wavelet has
the best representation in both time and frequency because
it is based on the Gaussian window. The Gaussian function
guarantees a minimum time-bandwidth product, providing
for maximum concentration in both time and frequency
domains. This is the best compromise for a simultaneous
localization in both time and frequency as the Gaussian
function’s Fourier transform is simply a scaled version of
its time domain function. In order to evaluate algorithm’s
quality, first we implement 16384-point FFT (best fit signal’s
length). Figure 14 shows spectrum plot.
We can see the fundamental frequency of this signal is
9.56 Hz. To evaluate this method, we apply the algorithm
with 4 different wavelet functions: Haar, Morlet, Daubechies

DWT with db4 on hardware. The coefficients of db4’s scale
function are:
√
√
√
√
3+ 3
3− 3
1− 3
1+ 3
√ ; h1 =
√ ; h2 =
√ ; h3 =
√
h0 =
4 2
4 2
4 2
4 2
(15)
The wavelet function coefficients are:
g0 = h3 ; g1 = −h2 ; g2 = h1 ; g3 = −h0

Fig. 14. Spectrum plot of PPG signal.

4 and Coiflet 1. Figure 15 is the CWT result with the Morlet
mother wavelet.

(16)

Each step of the wavelet transform applies the scale
function and wavelet function to the input data. If the original
data set has N values, the wavelet function will be applied to
calculate N/2 smoothed values and N/2 differences (reflecting
change in the data). In the forward wavelet transform, the
smoothed values are stored in the lower half of the N element
input vector and the differences are stored in the upper half
of the N element input vector.
The scaling and wavelet functions are calculated by taking
the inner product of the coefficients and four data values. The
equations are shown in equation (17) and equation (18).
Scaling function:
ai = h0 s2i + h1 s2i+1 + h2 s2i+2 + h3 s2i+3

(17)

Wavelets function:
ci = g0 s2i + g1 s2i+1 + g2 s2i+2 + g3 s2i+3

Fig. 15. Continuous Wavelet Transform visualization.

The area with lighter color represents large coefficients
area. From the flowchart and algorithm, max coefficient
corresponds to scale value 88.2, and so we get frequency
9.43 Hz. We can see that with the Morlet mother wavelet,
the Wavelets based approach gives a similar result to the
Fourier based one. Table IV shows the results of the other
mother Wavelet as well as the other signals.

(18)

Each iteration in the wavelet transform step calculates a
scaling function value and a wavelet function value. The
index i is incremented by two with each iteration, and new
scaling and wavelet function values are calculated. In the case
of the forward transform, with a finite data set (as opposed
to the mathematician’s imaginary infinite data set), i will be
incremented until it is equal to N-2. In the last iteration the
inner product will be calculated from calculated from s[N2], s[N-1], s[N] and s[N+1]. Since s[N] and s[N+1] don’t
exist (they are beyond the end of the array), this presents
a problem. This is shown in the transform matrix below
(figure 16). Note that this problem does not exist for the

TABLE IV
R ESULT OF A PPLYING A LGORITHM TO S OME P PG S IGNALS
No.

FFT

Morlet

Haar

Db4

Coif1

1

10.56

10.43

11.93

9.64

11.17

2

9.56

9.40

10.21

8.60

9.90

3

11.25

11.39

11.93

9.72

11.22

4

9.68

9.78

20.4

7.45

8.20

5

8.75

8.72

9.94

8.70

9.60

Based on the tables, different wavelet functions produces
different results compared to the Fourier transform. The
Morlet function has the smallest error (less than 2%), these
results can be accepted, but the Haar function has the largest
error value, so it isn’t suitable for signal processing. Other
functions, such as db4 and Coif 1, have the error less than
the Haar function but it is still not acceptable.
V. S IGNAL D ENOISING U SING DWT ON H ARDWARE
DWT with db4 mother Wavelet has proved its effectiveness. Noise is basicly removed from the original signal.
In this section, we will introduce a method to implement

Fig. 16. Forward transform matrix for an 8 element signal.

Haar wavelet, since it is calculated on only two elements,
s[i] and s[i+1]. A similar problem exists in the case of the
inverse transform. Here the inverse transform coefficients
extend beyond the beginning of the data, where the first
two inverse values are calculated from s[-2], s[-1], s[0] and
s[1]. This is shown in the inverse transform matrix below
(figure 17).
Three methods for handling the edge problem:
• Treat the data set as if it is periodic. The beginning of
the data sequence repeats rolling the end of the sequence
(in the case of the forward transform) and the end of
the data wraps around to the beginning (in the case of
the inverse transform).

Fig. 17. Inverse transform matrix for an 8 element signal.

Treat the data set as if it is mirrored at the ends. This
means that the data is reflected from each end, as if a
mirror were held up to each end of the data sequence.
• Gram-Schmidt orthogonalization. Gram-Schmidt orthogonalization calculates special scaling and wavelet
functions that are applied at the start and end of the
data set.
Zeros can also be used to fill in for the missing elements,
but this can introduce significant error. The algorithm published in this paper treats the data as if it were periodic.
After building DWT and IDWT algorithm, the next thing
that we have to do is to build the de-noising algorithm on
hardware. First of all, we need to estimate the threshold used
for de-noising. Using Donoho’s method as discussed in the
previous section, now we need to build a function to calculate
median. According to the median algorithm, firstly input data
sequence x with data length N are arranged from small to
large. Then:
•

 n
n


 x[ 2 ] + x[ 2 − 1]
if N is even
(19)
median(x) =
2

n

x[ ]
if N is odd
2
After calculating the median, using equation (13) and
equation (14) to calculate the threshold value. After determining the threshold, the samples of signals whose absolute
values are less than the threshold are set to 0 (hard threshold
method). The remaining values will be preserved.
VI. S YSTEM D ESIGN
A. Hardware Implementation

amplitude is very small. Thus it needs to be amplified and
converted to a voltage signal so that it can be measured by
MCU’s Analog-to-Digital module (ADC). This amplifying
and converting module is called transimpedance (TIA). In
practice, we can use the IC OPT101 by Texas Instruments.
This IC already contains a TIA module and a photodiode to
detect light signal.
The signal after photo-detector contains lots of other
undesired frequency components. Thus, a band-pass filter is
needed to remove the undesired signal, leaving the frequency
which is desired. People’s heart rate is from around 40 - 160.
The filter’s pass band should be in this range. The receiver
filter is set 0.8 Hz to 3 Hz as the pass band. In addition, the
filter should provide some gain added to the signal, because
even though TIA is put behind photo-detector, the overall
gain is not enough. Figure 19 shows a proposed design
of the filter and amplifier stage. Using Multisim Electronic
Workbench to simulate this design, the actual pass band is
0.714 - 3.2 Hz, the total gain is about 38 dB (it means the
PPG signal will be amplified up to 80 times).

Fig. 19. Schematic of the filter and amplifier stage.

The most important module of this design is the MCU.
The more powerful the MCU is, the higher efficiency we
will get. Nowadays, there are plenty of MCUs from many
manufacturers, such as TI, Atmel, Microchip, STMicroelectronics In this paper, An ARM Cortex-M4 processor-based
MCU is used. It’s called SAM4E16E. For convenience, this
MCU is integrated to an evaluation board called SAM4EEK, which is shown in figure 20. The language used for
programming is C/C++. We use Iar Embedded Workbench
as an IDE to compile code and download to board.

To calculate the heart rate continuously, a sensor is designed to acquire the PPG signal. The proposed design of
the hardware is shown in figure 18.

Fig. 18. Proposed hardware design.

The transmitter is a LED emitting light in the near IR
whose wavelength is 940 nm. The LED’s intensity is driven
by a Digital-to-Analog module (DAC) of the microcontroller
(MCU). After the transmitted infrared signal passes through
the test site (usually a reasonably translucent area with good
blood flow, such as a finger or earlobe), it is received at
the photo-detector. However, this is a current signal and its

Fig. 20. SAM4E evaluation board.

MCU will process the PPG signal to calculate the heart
beat value. This heart beat value will be sent to PC via UART
and displayed in a Graphic User Interface (GUI).

B. Software Implementation
1) Board Programming: As discussed in the theory section, the signal received from the photo-detector and the
human heart beat is congruent. It means every time your
heart beats, it generates a pulse signal. Although the signal
received from the photo-detector is weak, but after the block
of filtering and amplification, the signal is amplified many
times and can be measured by the ADC on board SAM4EEK. Heart rate is the number of beats in 60 seconds, but
if you read all data in the 60s, it will take a lot of time. A
more simple way, because heart rate usually does not change
continuously, so we can just read the ADC data in 10s, then
use FFT to calculate signal’s frequency multiply with 60,
we will get your heart rate parameters. An alternative way
to calculate heart beat is using the Timer Counter function
of MCU. Signal passing through the filter and amplifier will
become a pulse, so it can be counted by Timer Counter. This
result is then multiply with 6 only. Both FFT and Timer
Counter methods have their own advantages and disadvantages. FFT method requires a large memory, because it needs
at least 8192 points to give an acceptable resolution (the
suggestion is 16384 points). However, the higher resolution
is, the higher accurate the result is. With Timer Counter
method, memory requirement is not a problem, but the result
is less accurate than FFT. Figure 21 shows the flowchart of
the main program.

Fig. 21. Main program flowchart.

While doing delay routine, the Interrupt Service Routine
(ISR) will execute. In the ISR, ADC module will read the
converted result and save to the memory; or the Timer
Counter will read the number of pulse appearing if we use
the second method. Figure 22 shows the flowchart of the
ISR. A real device has been used to check the accuracy of
this design. This device is manufactured by OMRON. Some
tests were made on some people using both this design and
OMRON device. Table V shows the results of these tests.
As we can see, the difference between them is not really
significant. Instead of buying an expensive device, we can
use this design to get an acceptable result.
2) Designing PC Program: As discussed above, the heart
rate value after being calculated will be sent to PC and
displayed in a GUI. The GUI can be created by various of
language such as Java, C/C++, C#... In this paper, we use

Fig. 22. ISR program flowchart.
TABLE V
P ROPOSED D ESIGN VS O MRON D EVICE
No.

Proposed design

OMRON device

Error (%)

1

81

80

1.25

2

73

73

0.00

3

66

67

1.49

4

50

52

3.85

5

77

77

0.00

6

55

56

1.79

7

61

61

0.00

8

69

69

0.00

9

90

88

2.27

10

70

71

1.41

Java because it is a powerful language, with high performance and security. More over, a Java program just needs
to be written once on a platform (like Windows, Unix...),
and when we bring it to another platform, it still works
well. First of all, we need to build a connection between
PC and board. One of the most common interface used on
MCU is UART. SAM4E-EK has already integrated a RS232
port. The advantages of this interface are simplicity and
convenience. Therefore in this paper, we use UART as the
interface between PC and board. In Java, the library for this
interface is called Java Communication API (also known as
javax.com). However, this library supports Linux and Solaris
platforms only. To be able to use this interface in Windows,
it is necessary to install a third-party library called RXTX.
This library is available at www.rxtx.qbang.org. Figure 23
is the GUI of the UART setting. On this GUI, the user can
select the COM port to communicate as well as install the
parameters of the UART interface such as Stop bits, Flow
Control, Data bits ...

Fig. 23. UART interface setting.

The next thing is the main GUI. In this main GUI, the
health information such as heart rate, blood pressure and

blood glucose will be displayed. Figure 24 shows the main
GUI that we have designed.

Fig. 24. The main GUI.

In the system that we’re building, each time a user wants
to update their health information to the server, they need
to provide their personal information. So we need to build
a GUI to enter information. This GUI can be constructed as
shown in figure 25.

VII. C ONCLUSION
This paper has focused on researching methods of determining the health parameters such as the heart rate, the
blood pressure, the blood glucose by non-invasive, using the
algorithms to analyze the PPG signal to determine the desired
parameters. In the other hand, we have also researched the
theory and application of Wavelets in Biomedical Signal
Processing, they are the use of CWT to determine the value
of heart rate and DWT to remove noise from original PPG
signals. The denoising method based on DWT has been also
implemented on hardware. In addition, a personal health
information system has been designed. This system contains
a circuit to measure heart rate value of humans with the
support of SAM4E-EK board, a software to display and
transmit those values over the Internet, a website system
to display measured results for remote monitoring. In the
future, we plan to research some other methods to determine
more health parameters and develop this system to give users
more advantages, such as a mobile-based health monitoring
system.
R EFERENCES

Fig. 25. The GUI for entering information.

The final function of the PC program that we need to
build is the connection to Internet, because we need to
upload the information such as health parameters, personal
information... to the web-server. In Java, the java.net library
provides the methods to access to the Internet. After being
constructed, this program will continuously read data via
UART interface and show up to the GUI. Each time user
fills in their information and presses the button “Update”,
data will be sent via Internet to the web server.
3) Designing Web Page: There are plenty of language that
we can use to design a web page. In this paper, we choose
ASP.NET to create our web page. The reason for choosing
ASP.NET because it is a powerful platform supported by
Microsoft, with a simple programming model, supporting
multi-language, high security ability, convenient for web
application configuration. Figure 26 is the designed web
page.

Fig. 26. Web page for monitoring health information.

[1] J. Allen, “Photoplethysmography and its application in clinical physiological measurement,” Physiological Measurement (United Kingdom),
vol. 28, pp. 1–39, 2007.
[2] P. H. A. M. Z. D. X. N. B. K. X.P. Nguyen, R. Kronemayer and
N. Gretz, “Validation of a new non-invasive blood pressure measurement method on mice via pulse wave propagation time measurement
on a cuff,” Biomed Tech Journal (Berlin), vol. 56, pp. 153–158, 2011.
[3] M. L. P. T. E. M. Hartley CJ, Taffet GE, “Noninvasive determination
of pulse-wave velocity in mice,” Journal of Physiology - Heart and
Circulatory Physiology (USA), vol. 273, pp. 494–500, 1997.
[4] G. W. Langewouters GJ, Wesseling KH, “The static elastic properties
of 45 human thoracic and 20 abdominal aortas in vitro and the
parameters of a new model,” Journal of Biomechanics (USA), vol. 17,
pp. 425–435, 1984.
[5] H. A. Bramwell J., “The velocity of the pulse wave in man,” Proceedings of the Royal Society of London. Series B, Containing Papers of
a Biological Character (1905-1934), vol. 93, pp. 298–306, 1922.
[6] S. M. K. A. Yamakoshi KI, Shimazu H, “New oscillometric method
for indirect measurement of systolic and mean arterial pressure in the
human finger,” Medical Biological Engineering Computing Journal
(Heidelberg, Germany), vol. 20, pp. 307–313, 1982.
[7] A. M. A. Amerov Airat K., Chen Jun, “Molar absorptivities of glucose
and other biological molecules in aqueous solutions over the first
overtone and combination regions of the near infrared spectrum,”
Proceedings of the Royal Society of London. Series B, Containing
Papers of a Biological Character (1905-1934), vol. 58, pp. 1195–
1204, 2004.
[8] S. H. K. J. Jeon, I. D. Hwang and G. Yoon, “Comparison between
transmittance and reflectance measurements in glucose determinationusing near infrared spectroscopy,” Journal of Biomedical Optics
(USA), vol. 11, pp. 14–22, 2006.
[9] E. V. V. A. I. Pavlyuchko and L. A. Gribovb, “Calculations of molecular ir spectra in the overtone and combination frequency regions,”
Journal of Applied Spectroscopy (USA), vol. 78, pp. 639–645, 2011.
[10] K. B. C. S. Ramasahayam S., Haindavi K.S., “Non invasive estimation of blood glucose using near infra red spectroscopy and double
regression analysis,” Seventh International Conference on Sensing
Technology (New Zealand), pp. 627–631, 2013.
[11] Y. T. T. T. Y. K. K. Yamakoshi, Ogawa M, “Multivariate regression
and discreminant calibration models for a novel optical non-invasive
blood glucose measurement method named pulse glucometry,” Conf
Proc IEEE Eng Med Biol Soc (USA), vol. 126, 2009.
[12] Y. Y. Yamakoshi K., “Pulse glucometry: A new approach for noninvasive blood glucose measurement using instantaneous differential
near infrared spectrophotometry,” Journal of Biomed Optics (USA),
vol. 11, 2006.
[13] M. Y. P. R. Bhandare, P., “Multivariate determination of glucose in
whole blood using partial least-squares and artificial neural networks
based on mid-infrared spectroscopy,” Journal of Applied Spectroscopy
(USA), vol. 47, pp. 1214–1221, 1993.
[14] M. W. B. L. K. G. Smits, J.R.M., “Using artificial neural net- works
for solving chemical problems. part 1. multi-layer feed-forward networks,” Journal of Chemometrics and Intelligent Laboratory Systems
(Netherlands), vol. 22, pp. 165–189, 1994.