FETAL ELECTROCARDIOGRAM EXTRACTION AND

ANALYSIS
Thesis submitted in partial fulfillment
for the award of the degree of
Bachelor of Technology
In
Electronics and Communication Engineering
Submitted by
AJIT KUMAR SAHU (110EC0180)
JYOTIRANJAN RAJ (110EI0052)
Under the guidance of
Dr. Samit Ari
Assistant Professor

Department of Electronics & Communication Engineering

National Institute of Technology, Rourkela
Odisha-769008
INDIA
i

DECLARATION

I hereby declare that the work presented in the thesis entitled Fetal
Electrocardiogram Extraction And Analysis is a bonafide record of the
research work done by me under the supervision of Dr. Samit Ari,
Department of Electronics & Communication Engineering, National
Institute of Technology, Rourkela, India and that this thesis work has not
been presented for the award of any other degree.

AJIT KUMAR SAHU

JYOTIRANJAN RAJ
Dept. of Electronics & Comm. Engg.
National Institute of Technology,Rourkela
Odisha-769008
INDIA
ii

CERTIFICATE
Certified that this project thesis on FETAL ELECTROCARDIOGRAM
EXTRACTION AND ANALYSIS is a bonafide work of AJIT KUMAR SAHU

and JYOTIRANJAN RAJ who carried out the research project

under my supervision and guidance during Aug 2013-May
2014(7th & 8th Semester). This thesis has not been
submitted for any degree or academic award elsewhere.

Date :

Dr. Samit Ari

Place:
Assistant Professor
Dept. of Electronics & Comm. Engg.
National Institute of Technology, Rourkela
Odisha-769 008
INDIA

iii

ACKNOWLEDGEMENT

Firstly, we would like to express our deep sense of respect and gratitude towards our advisor and
guide Prof. Samit Ari, who has been the guiding force behind this work. We want to thank him for
introducing us to the field of Signal Processing and giving us the opportunity to work under him. It is
our good fortune to have got an opportunity to work under such a wonderful person. We sincerely

appreciate his inspiration; support and guidance. He has been instrumental in making this project
a success.
We also express our sincere gratitude to Prof S. Meher, Head of Department, Electronics and
Communication Engineering for allowing access to all valuable facilities which were a necessity
for us. We would also like to convey our gratitude to all the research scholars in our department
who assisted us during the completion of this project work and their well-timed and valuable
guidance that helped us to finish the work in the predetermined period of time.
We are obliged to staff members of department of Electronics and Communication Engineering,
for the valuable information provided by them in their respective fields. We are grateful for their
cooperation during the period of our assignment.
Our thanks and appreciations also go to our friends and batch mates in developing the project
and people who have willingly helped us out with their abilities.

Ajit Kumar Sahu

Jyotiranjan Raj

110EC0180

110EI0052
iv

ABSTRACT

Diagnosis of mothers and childs heart beat is very necessary during pregnancy and hence we
use Fetal electrocardiogram (FECG) extraction for the same. The signal contains precise
informations that can help doctors during pregnancy and labor. In this thesis, an easy-to use
method has been implemented using adaptive noise canceller (ANC). Using the ANC, an
effective algorithm has been proposed. The algorithm uses ANC, Least Mean Square (LMS)
method and a Simulink model for the extraction of FECG. The FECG extraction method has
been implemented using an algorithm implemented on MATLAB using Simulink models. The
extracted FECG signal is a noise free signal. The QRS complex has been detected using another
algorithm that counts the R-R peaks. The simulation result shows that heart rate of the FECG
signal can be counted using the detection algorithm.
This project evokes a complete model of the FECG extraction with the implementation of
effective algorithms and adaptive filters and finally gives the heart rate of the FECG signal.

Keywords:
Electrocardiogram, Adaptive Noise Cancellation, Maternal Heart Beat, LMS Algorithm, QRS
Complex, Peak Detection

CONTENTS
Declaration.ii
Certificate..iii
Acknowledgement.iv
Abstract...v
List of Figures.viii
1. INTRODUCTION.1
1.1 Extraction...2
1.2 Fetal Heart..3
1.3 Heart Signal...4
1.4 Literature Survey...5
1.5 Objective5
1.6 Database.6
1.7 Thesis Outline6
2. EXTRACTION OF THE FETAL ECG SIGNAL......8
2.1 Methodology....11
2.2 Algorithm.12
2.3 Proposed Simulink Model...13
2.4 Least Mean Square Algorithm.14
2.5 Implementation of the Algorithm15

vi

2.6 Experimental Results...19

2.7 Conclusion...22
3

QRS COMPLEX23
3.1 Methodology....25
3.2 Experimental Results.......28
3.3 Conclusion...29

CONCLUSION AND FUTURE WORKS.......30

4.1 Conclusions..30
4.2 Future Works...30

REFERENCES........32

vii

LIST OF FIGURES
Fig 1.1 Measurement of Fetal ECG.3
Fig 2.1 FECG Extraction Algorithm..12
Fig 2.2 Simulink Model 1..............................................13
Fig 2.3 Simulink Model 2..14
Fig 2.4 Maternal Heart Beat Signal...16
Fig 2.5 Fetal ECG Signal...16
Fig 2.6 Measured Signal .......17
Fig 2.7 Adaptive Noise Cancellation System........18
Fig 2.8 Measured Fetal ECG Signal..20
Fig 2.9 Output of the Adaptive Noise Canceller (ANC)...21
Fig 2.10 Peak Detection.21
Fig 3.1 QRS waveform......24
Fig 3.2 Original ECG signal..26
Fig 3.3 First Order Derivative having a smooth and rectified signal26
Fig 3.4 Second Order Derivative having a smooth and rectified signal26
Fig 3.5 Sum of First and Second Order Derivative...26
Fig 3.6 Square Pulse Output for each QRS complex26
Fig 3.7 Differentiation Technique implemented on MATLAB.................................................28
Fig 3.8 QRS Peaks.....29
viii

CHAPTER 1

INTRODUCTION

1.1 FETAL ECG EXTRACTION

Every year, one out of hundred babies born with some form of heart defects. This occurs due to
genetic syndrome, inherited disorder or environmental factors like misuse of drugs. In any case
regular monitoring of heart of the baby is necessary before it is born. Hence, Fetal ECG (FECG)
signals are necessary to monitor the heart condition of the baby, so that if any abnormalities
found then it can be solved clinically by the concerned doctors [1].
Fetal ECG monitoring is a widely used technique for diagnosis and to find out fetal
abnormalities. The physician can easily prepare himself/herself for the possible abnormalities in
a fetus by taking diagnosis of the fetal ECG signal during the pregnancy stage. Its the simplest
non-invasive method to diagnose various heart diseases. The various electrical activity of the
heart is represented by the Fetal ECG (FECG) and hence it provides valuable information about
its physiological state. The FECG signal can be easily obtained from the abdomen of a pregnant
women and the chest gives the maternal electrocardiogram (MECG) signal. The addition of the
MECG signal with the FECG signal is typically annoying one. By placing electrodes on the
maternal abdomen, the FECG signal, thus generated, gives minute details about the fetal
condition which are very useful during diagnosis. The abdominal ECG signal contains various
unwanted interferences, maternal ECG (MECG) signal and electromyogram (EMG) and the
FECG signal is corrupted by various noise and the skin impedance [2].
The electrical activity of the heart is described by the ECG. The ECG signal consist of three
types of wave. The details about QRS complex have been discussed in Chapter 2. The peaks of
the QRS complex give the information about the heart rate of the abdominal ECG (AECG)
signal. Hence, it is very crucial for the doctors to detect the heart defects before it causes any
damage to the fetus or the mother. When the ECG signal is taken from the abdominal leads, a
composite signal is picked [3].

There are various methods to extract FECG like wavelet transform, Doppler ultrasound, adaptive
filtering, correlation methods, blind source separation technique and a combination of blind
source separation methods and wavelet analysis. The heart rate of the FECG can be determined
by calculating the R-R peaks of the QRS complex [4]. But, as the FECG signal is merged with
MECG signal and various interferences, it is very difficult to calculate the heart rate from the
raw signal. Hence, the FECG is extracted from the raw signal to get the proper heart rate of the
fetal signal.
1.2 HEART OF THE FETUS
There are various organs that grows slowly inside a fetus. Amongst that aone of them is the fetal
heart. It goes through various changes with time during the early stages of pregnancy. After 22
days of pregnancy the heart states beating. From 18th - 20th week of conception the FECG signals
can be measured from the abdomen of the mother. In the maternal abdomen the skin and the fat
have very conductivity. Combining them forms a volume conductor which helps in the
propagation of the ECG signals to the surface. Fig 1.2 shows the measurement of ECG signal
from the fetus.

Fig 1.1 Measurement of Fetal ECG [5]

1.3 HEART SIGNAL

Action Potential
It is the membrane potential of the cell. When an electrical current is stimulated to a cell then the
cell goes through various mechanical contraction to give a potential known as action potential
[6].
Resting Potential
It is the static membrane potential of the cell. When the cells are at rest the membrane readily
allows ions like K+ and Cl- to pass into them because they are permeable membranes. The
outside of cell is positively charged than the inside of the cell. Hence, K+ ions flows into the cell
to balance the charge. Hence, a potential difference gets developed between the inside and
outside of the cell at the state of equilibrium. A resting cell is a polarised cell [7].
Depolarisation
Initially the cell doesnt allow Na+ ions inside due to higher permeability of the K+ ions. But
when the cell is excited it starts allowing the Na+ ions to move inside. This movement of Na+
ions contain the ionic currents. Hence, the membrane barrier reduces to Na + ions. The rushing of
Na+ ions into the cell is called Avalanche effect. Due to higher concentration of the K + ions in
the cell it tries to move out of the cell but cant move as fast as the Na + ions. Hence, the inside of
the cell becomes positive with respect to the outside of the cell due to imbalance of the K+ ions.
This process is called depolarisation [8].
Repolarisation
The repolarisation of cells depends on the time dependent and voltage dependent of the
membrane permeability changes for K+ ions compared with that for the Na+ ions. During

depolarisation the Na+ ions moment get slower and hence repolarisation occurs due to the
increase in moment of the K+ ions at the end of depolarisation [9].
1.4 LITERATURE SURVEY
Seeing a typical fetal ECG (FECG) signal, the doctor rectifies the fetal heart abnormalities. The
ECG device extract the FECG signal from the Abdominal ECG (AECG) signal and shows them
on the monitor. The heart rate of the fetal heart comes to be greater than the maternal heart rate.
The motivation behind this project are challenges like the extraction of a low amplitude and high
frequency signal from the high amplitude and low frequency signal. In this work, we have
ensured extraction of fetal ECG signal but we havent processed it in an efficient manner as done
in research work by Prasanth K and Baby Paul [10]. We have taken 4000 samples of the AECG
signal with peak amplitude of 3.5 millivolts. And then the fetal ECG signal rate comes in the
range 120-160 heart beats per minute. . And hence the fetal heart beats faster than the maternal
heart and is very sensitive to heart abnormalities.
Our project ensures the peaks of the FECG is detected properly but not as efficiently done by
Butler Pyke [11], it uses Pan-Tompkins algorithm for the detection of the QRS complex. In this
project, we have used the differentiation technique for the detection of the R-R peaks in a QRS
complex. Though the method is not so efficient, it gives us the required output as i.e. almost
accurate FECG heart rate.
1.5 OBJECTIVE
Two ECG signals were taken; one from the chest of the mother and another from the abdomen of
the mother that contains both the fetus heartbeat signal and maternal heartbeat. The objective of
our study is to separate both these parts and thus extract the fetus heartbeat signal and then detect
the R-peaks in order to determine the fetus heart rate.
i) To extract the fetus heartbeat signal from the ECG signal taken from the abdomen.
5

 Regenerating the waveforms of both ECG signals in MATLAB using the data taken from
the ECG machine.
Separating both signals and reducing any noise present in the signal to obtain the fetus
heartbeat signal using ANC.
ii) To count the R-peaks and calculating the heart rate of the fetus by using the R-R separation.
Applying a differentiation technique to the QRS-complex to differentiate the extracted
signal up to 3rd order to get the peaks.
Setting a threshold and comparing the peaks with it to detect the R-peaks.
Finally counting the R-peaks per one minute to obtain the heart rate of the baby.
1.6 DATABASE
Abdominal and direct fetal electrocardiogram database included 38 and 41 weeks of gestation
between research materials, labor derived from 5 different girls multichannel fetal
electrocardiogram (FECG) has records. Entries fetal electrocardiogram (Itam company, Zabrze,
Poland), Department of Obstetrics, Medical University of Silesia, through acquisition and
analysis system was purchased KOMPOREL. Each record in the stomach and the fetal head to
the maternal fetal electrocardiogram obtained directly from the record contains four different
signals.
1.7 THESIS OUTLINE
Chapter 1 of the thesis explains the background of the fetal electrocardiogram extraction and
also give details about literature survey on FECG extraction. Next it gives detail about the online
data used in this thesis and the experimental setup to implement the extraction process and get
the FECG signal as output.

Chapter 2 of the thesis explains the algorithm of fetal ECG extraction. Steps that has been taken
and the database that has been used to extract the FECG signal from the Abdominal ECG
(AECG) signal. Then the algorithm is implemented on MATLAB and the output is taken as
plots.
Chapter 3 explains the method used for the detection of R-R peaks from the QRS complex of
the FECG signal.
Chapter 4 describes the extraction output which is final output of the whole thesis work and
gives the concluding remark and future work.

CHAPTER 2

EXTRACTION OF THE
FETAL ECG SIGNAL

Several methods are used for the extraction of fetal ECG signal from maternal ECG signal which
is actually a mixture of both mother and fetus heartbeat signals. Some of the most widely used
methods are Principal Component Analysis (PCA), Independent Component Analysis (ICA) and
Adaptive Noise Canceller (ANC). Both PCA and ICA methods come under a category known as
Blind Source Separation (BSS) which is due to the lack of any prior knowledge about the source
or mixing type. The above mentioned methods are briefly described below.
ANC (Adaptive Noise Canceller)
This method is used for non-stationary type noise or interference which is not necessarily a
random process. Adaptive filters are systems with a linear filter that uses a transfer function
which is controlled by different variable parameters and these parameters can be adjusted
according to an optimization algorithm [12]. Most adaptive filters are digital filters owing to the
complexity of the optimization algorithms used. In case of an ANC no info about the signal and
noise characteristic is available and both the noise and signal are uncorrelated. Another reference
signal is obtained from a second source which is strongly correlated with the noise but
uncorrelated with the signal. When the signal and noise are stationary, an adaptive filter acts as a
fixed filter whereas it acts as a notch filter or comb filter for a periodic interference. Least Mean
Squares (LMS) and Recursive Least Squares (RLS) filters are a few examples of adaptive filters
and all these filters described above can only be applicable for additive noise.
PCA (Principal Component Analysis)
In this method the axis is rotated corresponding to the direction of maximum covariance in order
to encode the second-order dependencies but the high order dependencies in the data are not
addressed by it. Basically PCA can be termed as a statistical procedure to obtain a set of linearly
uncorrelated variables called principal components from a set of observations of possibly
9

correlated variables using some kind of orthogonal transformation technique. In other words it
can be described as an orthogonal linear transformation used to transform the data into a new
coordinate system in which the greatest variance of the data lies on the first coordinate otherwise
known as the first principal component and the second greatest variance on the second principal
component and so on. The dimensionality in a data can be reduced while preserving the
characteristics contributing the most to its variance by eliminating the higher principal
components using PCA [13].
ICA (Independent Component Analysis)
This technique is used to recover a set of independent signals from the measured signals or it can
also be termed as a computational method that separates multivariate signals into a number of
additive subcomponents [14]. Here the number of independent signals obtained is same as the
number of measured signals where each measured signal is represented as a linear combination
of the independent signals obtained from it.
Mathematically,
Yi = k1X1 + k2X2 + + knXn

(2.1)

Here Yi is the ith measured signal and X i is the ith independent signal derived from the set of
measured signals and hence the entire system can be expressed as,
Y = KX

(2.2)

INC is perhaps the most widely used statistical technique used to deal with BSS problems which
decomposes a multivariate signal into non-Gaussian independent signals. Maximum Entropy
Method is an example of ICA [15] [16].

10

Out of the three methods stated above the Adaptive Noise Canceller (ANC) method is going to
be used in this thesis. Hence, the proposed fetal ECG signal extraction process is described in the
next few topics as follows,
2.1 METHODOLOGY
The electrocardiogram (ECG) signal for both the mother and fetus are retrieved from the given
data that is sampled at 5000 Hz using a smoothing filter to give the discrete data a somewhat
smoother shape. The filter used here is a digital filter known as the Savitzky-Golay Filter. The
main purpose of using this filter is to smooth the signal by increasing the signal-to-noise ratio
(SNR) without distorting the signal significantly. This filter uses a convolution method to
achieve the goal. The heart rate of a fetus is significantly higher than that of the maternal heart
rate which is in this case approximately a rate of 85 beats per minute whereas for the fetus it is
approximately 132 beats per minute. In general, the fetus heart beats faster than that of the
mother, with a range of rates from 120 to 160 beats per minute. The fetal electrocardiogram
signal has a much weaker amplitude as compared to its maternal counterpart [17] [18]. For
example here the mother ECG signal has a peak of 3.5 millivolts whereas the fetal ECG signal
has a peak of just 0.25 millivolts. ECG signals are taken from two different locations of the
mothers body, the chest and the abdomen. The chest signal gives the original mother ECG
signal whereas the signal obtained from the abdomen is a mixture of both mother and fetus
heartbeat signals usually dominated by the maternal component propagated from the chest cavity
to the abdomen. A linear FIR filter can be used with 10 randomized coefficients in order to
describe this path. There may be some additional broadband interference associated with both
mother and fetus signals that can be eliminated by the addition of a small amount of uncorrelated
Gaussian noise. The task performed by the adaptive filter is to adaptively remove maternal

11

component from the fetus heartbeat signal. To do so it needs a reference signal which is nothing
but the signal generated from the maternal ECG itself. Like the fetal ECG signal the maternal
ECG signal is also expected to contain some additive broadband noise [19].
2.2 ALGORITHM USED FOR EXTRACTION
The extraction method described in this thesis is based on a MATLAB code that consists of 4
simple steps as described in the flowchart of the proposed algorithm shown below in Fig. 2.1.

GENERATE ECG SIGNALS IN MATLAB

FECG EXTRACTION USING ANC

RECOVERING THE FECG

COUNTING THE NUMBER OF PEAKS
TO OBTAIN THE HEART RATE
Fig 2.1 FECG Extraction Algorithm

12

2.3 PROPOSED SIMULINK MODEL

To identify the heart rate of the fetus based on sensor data from two electrodes a suitable
Simulink model is proposed. Both the primary signals i.e. the mothers and babys heart beat
signal have some superimposed noise associated with them. The goal of this model is to filter out
everything except the babys heartbeat and calculate the period of the signal. The proposed
Simulink model is shown below in Fig. 2.2 and Fig 2.3.

Fig 2.2 Simulink Model 1

13

Fig 2.3 Simulink Model 2

2.4 LEAST MEAN SQUARE (LMS) ALGORITHM
Calculate the error and find the least mean of it. This is what the algorithm does. The error signal
is the difference of the desired output signal and the actual output signal. For deriving this
algorithm, the steepest descent algorithm is used [20]. Now, what a steepest descent algorithm
does? It finds out the gradient vector by calculating the derivative of the error. It is a recursive
algorithm and helps in finding out the wiener filter.
Now, what LMS algorithm does? It just minimizes the [()2 ] i.e. the mean of the mean of the
square of the error. The LMS algorithm is based on the Steepest Descent (SD) algorithm but it is
somehow different because it estimates the value continuously.
The SD algorithm is a deterministic gradient method whereas the LMS algorithm is a stochastic
gradient method [21] [22].
Using LMS algorithm we cannot find the exact values and hence we cant compute the optimal
weights. The weight values are never updated to the optimal values. But it converges the mean of

14

the square of the error, hence, it changes the optimal weight although there is no changes in the
weights.
The weight update equation has been given below:
W+1 = []

(2.3)

2.5 IMPLEMENTATION OF THE ALGORITHM

The different steps proposed in the algorithm are briefly described below,
I. Generation of the ECG signals using MATLAB
MATLAB is an easy-to use tool which is very helpful in the extraction of the Fetal ECG (FECG)
signal from the Abdominal ECG (AECG). Using MATLAB we generate the signal on which the
task can be performed and implemented easily [23]. MATLAB contains a function SavitzkyGaloy filter function and using this command the required signals are generated.
MATLAB coding is used to simulate the shapes of the ECG signals for both the mother and the
baby. The maternal ECG signal generated from mothers heart using the given data in previous
section is shown in Fig. 2.4. The peak amplitude of this signal is 3.5 millivolts and the heart rate
being 85 beats per minute. But the fetus heart beats noticeably faster than that of the mothers
ranging between 120 to160 beats per minute as stated previously [24] [25]. The FECG signal
generated is shown in Fig. 2.5 and the measured signal taken from the mothers abdomen which
is usually dominated by the mothers heartbeat signal is shown in Fig. 2.6.

15

Fig 2.4 Measured ECG signal from the abdomen of the mother i.e. Abdominal ECG
(AECG) signal

Fig 2.5 Maternal Heartbeat Signal

16

The above signal shows the Maternal Heartbeat Signal being generated by using a SavitzkyGaloy filter. 4000 samples of the signal has been taken with peak voltage of 3.5 millivolts.

Fig 2.6 Fetal Electrocardiogram Signal

II. FECG Signal Extraction Methods Used

1) Adaptive Noise Canceller (ANC):
Two signals are needed to remove background artefacts including different noises or
interference already present in the FECG signal by using adaptive filtering. The first one is
the FECG signal added with MECG signal and the second being the reference signal which is
also the signal to be cancelled to get the FECG signal [26] [27]. The reference signal is
nothing but equal to the MECG signal itself. Noises present in both the signals must be wellcorrelated i.e. the noise in the primary and secondary signals. In fetal ECG extraction
procedure adaptive filters are used to adaptively remove a mothers heartbeat signal from the

17

fetal ECG signal to get the heartbeat signal of the baby or the fetus. When the operating
environment is stationary, these filters possess constant shape and orientation for the errorperformance surface. But for operation in an environment that is not stationary, the bottom of
this surface moves continually along with the chances of changing orientation and curvature.
Hence, for non-stationary inputs, the filter seeks the bottom of the error performance surface
along with continually tracking it down. The block diagram for a basic ANC system is shown
below in Fig. 2.7.

Fig 2.7 Adaptive Noise Cancellation System

By making exact measurements of the gradient vector at each iteration and choosing the
step-size parameter suitably, the tap-weight vector computed would converge to the
optimum Wiener solution using the method of steepest-descent. Several recursive
algorithms can be used to change the tap-weight vector after each iteration. Least Mean
Square (LMS) algorithm is one of such algorithms that has a certain advantage as it

18

doesnt includes any steps regarding matrix inversion [28] [29]. This can be stated
mathematically as,
Wk+1 = Wk + 2 k Xk

(2.4)

where Wk is the tap-weight vector, k is the error and Xk is the input signal vector at kth
iteration whereas is the step size parameter.

2) High Frequency Removal Using Digital Filter:

FIR and IIR filters are the two primary types of filters widely used in Digital Signal
Processing (DSP). Among those FIR filters can be easily designed in linear phase and
some of the calculations can be omitted by using FIR filters, thus providing some
important computational efficiency. In order to count the fetal heart rate the R-peaks of
the FECG signal must be extracted [30] [31]. An FIR filter with appropriate filter
coefficients is used to remove high frequencies in order to achieve this goal. Then a
threshold value is set and the peaks with higher values than the threshold are detected as
the R-peaks of the FECG signal. The number of R-peaks are counted and from the R-R
interval the fetal heart rate is calculated.

2.6 EXPERIMENTAL RESULTS

Heart rate is calculation of the beating of the heart. Heart beats in a fixed time duration and the
calculation of number of beats i.e. number of R peaks per minute gives the heart rate of the ECG
signal.

19

Finally the signal is cleaned up and a threshold level is set so that any value above it can be
considered as a peak and hence the number of peaks in the signal can be counted. The heart rate
counting has been performed using the QRS detection techniques that has been discussed in
Chapter 3. The most optimized technique has been to count the R peaks of the FECG signal.
Counting the R peaks the heartbeat of the Fetus is calculated. The heart beat comes to be 135
beats per minute. This is comes in the range exactly [32]. As the fetal heart rate is greater than
the maternal heart rate and it comes in the range 120-160 beats per minute, this experiment
verifies the same aptly. Finally, all the plots are shown below that explains the signal forms of
the fetal ECG (FECG) signal.
Heartbeat calculation is never a difficult task if the fetal signal has been extracted from the
abdominal ECG (AECG) signal. Because, the first step is always to extract the FECG signal and
then applying optimized QRS complex detection techniques heartbeat can be calculated.

Fig 2.8 Measured Fetal ECG (FECG) Signal

20

Fig 2.9 Output of the Adaptive Noise Canceller (ANC)

Fig 2.10 Peak Detection

21

2.7 CONCLUSION
In this chapter the fetal ECG (FECG) extraction algorithm has been implemented successfully
using the adaptive noise canceller (ANC). The algorithm is implemented on MATLAB and then
output is taken using different plots. Then the peaks were detected using the differentiation
technique which has been explained in the Chapter 2 of this thesis. Finally, the heartbeat of the
fetus is measured successfully.

22

CHAPTER 3

QRS COMPLEX
DETECTION

23

Now a days it has been a trend to simulate ECG signals using microcomputers and various signal
processing devices. According to a survey the processing of ECG signals using systems based on
microcomputers is extremely efficient. There are many systems which has already been designed
to perform such tasks on ECG signals. And various signal processing has been done on them to
know the completed details about the ECG signals. But the ECG devices require a precise
detection of the QRS complex so as to run a diagnosis and detect the abnormalities. Hence, QRS
peak detection is a very crucial part of ECG signal processing. The figure below shows a QRS
waveform [33].

Fig 3.1 QRS waveform [30]

The transfer of stimuli or signal to atria and ventricles and then the return of the same stimuli or
signal from the atria and ventricles is what forms an Electrocardiogram (ECG) signal. This signal
is called Electrical activity occurring in the muscles of a human being. Fig 3.1 shows some kind
of ECG wave which consists of various parts from P and it goes alphabetically to U. Details
about all the parts have been explained below.
P wave This part of the signal is generated due to the depolarisation of the atria. It is called
activation.

24

QRS Complex- This part of the signal is generated due to the depolarisation of the ventricles. It
is also called activation.
The other parts of the waveform forms the part of depolarisation or recovery. The parts include
ST wave, T wave and U wave [34].
Hence, for conclusion the PQRSTU wave represents the complete electrical activity of the heart
due to the atria and ventricles. It is one kind of a cycle process that is repeated again after the
recovery of the heart. As long as this process continues it gives us a continuous ECG signal
which also time varying.

3.1 METHODOLOGY
Differentiation Technique for QRS Detection
In a cardiac cycle, the QRS complex has the largest slope i.e. larger rate of change of voltage.
This is due to the rapid conduction and depolarisation characteristics of the ventricles. The
derivative gives the rate of change. So in an attempt to develop an algorithm to detect the QRS

complex the operation would be the most logical starting point.

There are various techniques like Pan-Tompkins model, bandpass filtering technique, template
matching technique etc. that can be implemented to detect a QRS wave. Amongst them the
differentiation technique is also an efficient technique detect the R peaks of a QRS complex. The
basis of many detection techniques is differentiation. It is basically a high pass filter. Whenever
the derivative of a signal is taken the higher frequencies characteristic of the QRS complex gets
amplified while attenuating the lower frequencies of the P and T wave [35].
An algorithm based on the first order, second order and third order derivative has been used to
detect the peaks. The figures below shows the steps involved in the signal processing.

25

Fig 3.2 Original ECG signal [22]

Fig 3.3 First Order Derivative having a smooth and rectified signal [22]

Fig 3.4 Second Order Derivative having a smooth and rectified signal [22]

Fig 3.5 Sum of First and Second Order Derivative [22]

Fig 3.6 Square Pulse Output for each QRS complex [22]
26

The absolute values of the first and second order derivative can be calculated from the ECG
signal as
0 () = | () ( 2) |

(3.1)

1 () = | () 2( 2) + ( 4) |

(3.2)

These two data buffers, 0 () and 1 (), are first scaled and then summed to give the
following output
2 () = 1.30 () + 1.11 ()

(3.3)

The data buffer 2 () is properly scanned until a certain threshold is met or exceeded.
2 () 1.0

(3.4)

Once this condition is met for a data point in 2 (), the next eight points are needed to
compared to the threshold. The segment might be part of the QRS complex. This algorithm
detects the pulse and in addition to that it has the advantage that it produces a pulse which is
proportional in width to the complex. It has a particular disadvantage that it is sensitive to highfrequency noise.

27

3.2 EXPERIMENTAL RESULTS

The following figures show the MATLAB implementation of the code with peak detection.

Fig 3.7 Differentiation Technique implemented on MATLAB

The above figure shows the first order, second order and cumulative order differentiation of an
ECG signal. The peaks in the first signal are not visible properly but once it reaches the
cumulative order differentiation the peaks are quite visible enough to find out the R peaks.

28

Fig 3.8 QRS Peaks

The above figure shows the QRS peaks being detected by thresholding and locating the
maximum which is then converted into beats. In one epoch, number beats in 12 second has been
calculated and thereby heart rate has been computed.
3.3 CONCLUSION
In this chapter a technique has been explained for the detection of R-R peaks from the QRS
complex. Hence, the algorithm for the technique has been implemented successfully. And then
the peaks of the FECG are detected and the heartbeat of the fetus is measured.

29

CHAPTER 4

CONCLUSIONS AND
FUTURE WORKS

30

4.1 CONCLUSIONS
The outputs of the extracted signal were recorded on MATLAB. Finally, the fetal ECG signal is
extracted and heartbeat of the signal is calculated. The algorithm based on Adaptive Noise
Canceller (ANC) is proposed and implemented successfully. The performance of the algorithm
has been verified successfully on MATLAB and the algorithm is found to be highly efficient. It
is found after successful implementation that fetal ECG (FECG) signal can be successfully
extracted by using Least Mean Square (LMS) algorithm for tap-weight vectors. The LMS
algorithm is implemented by MATLAB codes and hence the LMS algorithm implements ANC
successfully. R peaks were also detected successfully giving the final heart rate of the signal.

4.2 FUTURE WORKS

The results obtained matches with the actual FECG signals significantly and will suffice to prove
correct. The work done here is on-going and it has many scopes in future because it hasnt been
detailed yet. The future work would include designing an ECG with in-build software
implementing the above extraction algorithm. Then the technique can be used widely in product
development. But the extraction techniques may change with time and hence more efficient and
error free techniques will be developed in future. In this way the project can move forward in
future.

31

REFERENCES
[1] International Journal of Advanced Research in Electrical, Electronics and Instrumentation
Engineering Vol. 2, Issue 4, April 2013
[2] http://en.wikipedia.org/wiki/Electrocardiography last accessed on Jan, 2014
[3] http://zone.ni.com/reference/en-XX/help/372357A-01/lvaftconcepts/aft_lms_algorithms/ last
accessed on Feb, 2014
[4] http://en.wikipedia.org/wiki/Least_mean_squares_filter last accessed on Mar, 2014
[5] http://www.cs.cmu.edu/~aarti/pubs/ANC.pdf last accessed on April, 2014
[6] http://ijater.com/files/288abb76-cc40-4bdc-ba4b-b9f4772fab8c_ijater_10_14.pdf
[7] American National Standard for Ambulatory Electrocardiographs, publication ANSI/AAMI
EC38-1994, Association for the Advancement of Medical Instrumentation, 2004
[8] AHA ECG database. Available from Emergency Care Research Institute, 5200 Butler Pike,
Plymouth Meeting, PA 19462, 2010
[9] Fahim Sufi, Qiang Fang, Ibrahim Khalil, and Seedahmed S. Mahmoud, Novel Methods of
faster Cardiovascular Diagnosis In Wireless Telecardiology, IEEE Journal on Selected Areas In
Communications, Vol. 27, No. 4 2005
[10] Rachid Merzougui, Mohammed Feham, Algorithm of remote monitoring ECG using
mobile phone: Conception and implementation, Third International Conference on Broadband
Communications, Information Technology & Biomedical Applications. 2006
[11] Turker Ince, Serkan Kiranyaz, and Moncef Gabbouj, A Generic and Robust System for
Automated Patient-Specific Classification of ECG Signals, IEEE Trans. Biomed. Eng. vol.56,
pp.1415-1426 2007
[12] L. Biel, O. Pettersson, L. Philipson, and P. Wide.Ecg analysis: an approach in human
identification,.IEEE Trans. Instrum. Meas., 50(3):808812 2008

32

[13] X.Afonso, W.J. Tompkins, T.Nguyen, S.Luo, ECG beat detection using filter, IEEE
Trans. Biomed. Eng. vol. 46, pp. 230-236 Vikram Goyal, J2ME Tutorial, Part 1: Creating
MIDlets, 2002
[14] Vikram Goyal ,J2ME Tutorial, Part 2: User Interfaces with MIDP 2.0, 2001
.
[15] S. Akselrod, D. Gordon, F. A. Ubel, D. C. Shannon, A. C. Barger, and R. J. Cohen.Power
spectrum analysis of heart rate fluctuation: A quantitative probe of beat to beat cardiovascular
control. Science, 213(1981):220222 1999
[16] MIT-BIH Database distribution, Massachusetts Institute of Technology, 77 Massachusetts
Avenue, Cambridge, MA 02139, 1998. http://www.physionet.org/physiobank/database/mitdb/
2000
[17] R.M. Rangayyan,Biomedical Signal Analysis: A Case-study Approach, Wiley
Interscience, New York, 2001, pp.18-28. 2001
[18] P. S. Hamilton and W. J. Tompkins.Quantitative investigation of qrs detection rules using
the mit/bih arrhythmia database,Biomedical Engineering, IEEE Transactions on, BME33(12):11571165, 2004
[19] P. S. Hamilton and W. J. Tompkins.Quantitative investigation of qrs detection rules using
the mit/bih arrhythmia database,Biomedic, 2000
[20] Ahlstrom, M. L. and Tompkins, W. J. 1983. Automated high-speed analysis of Holter tapes
with microcomputers. IEEE Trans. Biomed. Eng., BME-30: 65157, 1999
[21] Ahlstrom, M. L. and Tompkins, W. J. 1985. Digital filters for real-time ECG signal
processing using microprocessors. IEEE Trans. Biomed. Eng., BME-32: 70813, 2006
[22] Balda R. A., Diller, G., Deardorff, E., Doue, J., and Hsieh, P. 1977. The HP ECG analysis
program. Trends in Computer-Processed Electrocardiograms. J. H. vanBemnel and J. L.
Willems, (eds.) Amsterdam, The Netherlands: North Holland, 197205, 2008
[23] Dobbs, S. E., Schmitt, N. M., Ozemek, H. S. 1984. QRS detection by template matching
using real-time correlation on a microcomputer. Journal of Clinical Engineering, 9: 197212,
2003

33

[24] Friesen, G. M., Jannett, T. C., Jadallah, M. A., Yates, S. L., Quint, S. R., Nagle, H. T. 1990.
A comparison of the noise sensitivity of nine QRS detection algorithms. IEEE Trans. Biomed.
Eng., BME-37: 8597, 2002
[25] V. Zarzoso and A. K. Nandi, Noninvasive fetal ECG extraction: blind separation versus
adaptive noise cancellation, IEEE Trans. on Biomed. Eng., vol. 48 (1), pp.12-18, Jan. 2001.
[26] Khamene, A and Negahdaripour, A New Method for the Extraction of Fetal ECG from the
Composite Abdominal Signal, IEEE Trans. Biomed. Engineering 47 2000.
[27] C.W. Li, C.X.Zheng, C.F.Tai, Detection of ECG Characteristic Points Using Wavelet
Transforms; IEEE Transaction on Biomedical Eng., 42.No.1:22-28, January 1995.
[28] P. P. Kanjilal, S. Palit, and G. Saha, Fetal ECG extraction from single-channel maternal
ECG using singular value decomposition, IEEE Transactions on Biomedical Engineering, Vol.
44, pp. 5159, 1997.
[29] T. Solum, I. Ingermarsson, and A. Nygren, The accuracy of abdominal ECG for fetal
electronic monitoring, Journal of Perinatal Medicine, Vol. 8, No 3, pp. 142149, 1980.
[30] K. Karlsso, H. Lilja, K. Lindecrantz, and K. G. Rosen, Microprocessor based waveform
analysis of the fetal electrocardiogram during labor, International Journal of Gynaecol and
Obstetrics, Vol. 30, No. 2, pp. 10916, 1989.
[31] M. B. I. Reaz and L. S. Wie, Adaptive linear neural network filter for fetal ECG
extraction, Proceedings of International Conference on Intelligent Sensing and Information
Processing, Chennai, India, pp. 321324, January 2004.
[32] MIT-BIH Database distribution, Massachusetts Institute of Technology, 77 Massachusetts
Avenue, Cambridge, MA 02139, 1998. http://www.physionet.org/physiobank/database/mitdb/
2000
[33] V. Zarzoso and A. K. Nandi, Noninvasive fetal ECG extraction: blind separation versus
adaptive noise cancellation, IEEE Trans. on Biomed. Eng., vol. 48 (1), pp.12-18, Jan. 2001.
[34]
http://zone.ni.com/reference/en-XX/help/372357A-01/lvaftconcepts/aft_lms_algorithms/
last accessed on Feb, 2014

34

[35] Friesen, G. M., Jannett, T. C., Jadallah, M. A., Yates, S. L., Quint, S. R., Nagle, H. T. 1990.
A comparison of the noise sensitivity of nine QRS detection algorithms. IEEE Trans. Biomed.
Eng., BME-37: 8597, 2002

35