[Downloaded free from http://www.jr.ietejournals.org on Sunday, July 04, 2010, IP: 123.238.23.

177]

A Comparative Analysis of De-Noising Algorithms

for Fetal Phonocardiographic Signals
Vijay S. Chourasia and A. K. Mittra1
Electronics Engineering Department, MIET, Kudwa, Gondia (M.S.), India.
Assistant Professor, Electronics Engineering Department, MIET, Gondia (M.S.), India.

ABSTRACT
This paper is aimed at the selection of de-noising algorithm for de-noising of the fetal phonocardiographic (fPCG)
signals. Fourier-based analyzing tools have some limitations concerning frequency and time resolutions. Although
wavelet transform (WT) overcomes these limitations, it requires selection of appropriate de-noising algorithm. The
universal threshold, minimax threshold and rigorous SURE (Steins Unbiased Risk Estimate) threshold algorithms
along with soft or hard thresholding rule have been compared for de-noising of these signals. The mean-squared error (MSE) is used to evaluate the performance of these algorithms. The results show that, the rigorous SURE threshold algorithm with soft thresholding rule has a better performance for the analysis of fPCG signals when using the
fourth-order Coiflets wavelet. The proposed approach is simple and proves to be effective when applied for the selection of de-noising algorithm for the fPCG signals. These de-noised signals can be used for the accurate determination
of fetal heart rate (FHR) and further diagnostic applications pertaining to the fetus.
Keywords:
De-noising algorithms, fetal phonocardiography, wavelets.

1.

INTRODUCTION

Phonocardiography is a continuous noninvasive,

low-cost and accurate monitoring method that takes
care of fetal well-being [1]. Using this method, longterm measurement of the FHR and collection of heart
sound artifacts becomes possible. The widely accepted
ultrasound Doppler process is not suitable for taking
measurements over extended periods of time [2]. The
main advantages of the fPCG technique are its passivity
(noninvasiveness) and simplicity [3].
Despite all these advantages of fPCG, this technique is
not popular with the obstetricians because of its poor
signal-to-noise ratio (SNR) at the time of recording [4].
The fPCG signals recorded from maternal abdominal
surface are contaminated by various unwanted signals,
like maternal organ sounds, fetal movement effects
and the ambient noise; hence this technique requires
robust signal-processing to extract the fetal heart sound
signals [5]. To overcome this limitation of poor SNR
and to exploit the advantages of fPCG technique for
fetal home-care applications, it is required to adopt
an effective de-noising method for the extraction of
accurate FHR from the fPCG signals. There have been
significant efforts to develop long-term monitoring
methods based on fPCG technique. All these methods
have projected different types of sensors and filtering
schemes, as well as different numbers of channels, for
this purpose.
10

The main difficulty in dealing with the fPCG signals is

their extreme variability and necessity to operate on a
case-to-case basis [6]. Another important aspect of these
signals is that the information of interest is often a combination of features that are well localized temporally or
spatially. This requires the use of analytical methods sufficiently versatile to handle events that can be at opposite
extremes in terms of their time-frequency localization.
Wavelet analysis has proved to be one of the most successful techniques for the analysis of signals at multiple
scales and has rendered many successful applications in
the area of biomedical signal processing [7].
The choice of wavelet family, mother wavelet and
de-noising algorithm greatly affects the accuracy of the
wavelet analysis of the signal. The presented work deals
with this problem in two steps first, by choosing the
appropriate de-noising algorithm from universal threshold, minimax threshold and rigorous SURE (Steins
Unbiased Risk Estimate) threshold along with soft or
hard thresholding rule for de-noising of the fPCG signals;
and second, by selecting the best wavelet family and
mother wavelet on the basis of the nature of the signal to
be analyzed and properties of the wavelet families.
The rest of the paper is organized as follows. In section 2
theory of wavelet transform is discussed. A brief description of the various wavelet families and de-noising algorithms are presented in section 3. Section 4 elaborates the
comparison and results. Finally, section 5 summarizes
IETE JOURNAL OF RESEARCH | Vol 55 | ISSUE 1 | JAN-FEB 2009

[Downloaded free from http://www.jr.ietejournals.org on Sunday, July 04, 2010, IP: 123.238.23.177]
Chourasia VS and Mittra AK: A Comparative Analysis of De-noising Algorithm

the conclusions drawn from the previous sections.

2.

THEORY OF WAVELET TRANSFORMS

In WT the time domain waveforms are mapped into a

frequency-time domain while preserving both frequency
and time information. The main idea of wavelet analysis
is to measure the degree of similarity between the original
waveform s(t) and the basic function of the WT, also called
the mother wavelet, through wavelet coefficients computation. The calculation process is performed on shifted version of the mother wavelet, thus moving along the time;
and on stretched or compressed version of the mother
wavelet, thus varying the frequency. The continuous
wavelet transform (CWT) is defined as the convolution
between the original signal s(t) and a wavelet a,b(t).
W (a,b) =

S(t)

ab

(t)dt


1 +
tb 
s(t)
dt
=

a
a

(1)

where s(t) is the input signal; a is the scaling factor; b

is the translation parameter; and (t) is the transforming
function, called mother wavelet. The wavelet function
is given by
a,b =

t b 

a a

(2)

The Discrete Wavelet Transform (DWT) coefficients are usually sampled from the CWT on a
dyadic grid, choosing parameters of translation
b = n*2m and scale a = 2m. The wavelet function in DWT
is defined as
m,n (t) =

1
2m

t n2m


2 m

(3)

DWT analyzes the signal by decomposing it into its

coarse and detail information, which is accomplished by
using successive high-pass and low-pass filtering operations, on the basis of the following equations:
yhigh(k)=nx(n).h(2kn)

(4)

ylow(k)=nx(n).g(2kn)

(5)

where yhigh (k) and ylow (k) are the outputs of the high-pass
and low-pass filters with impulse response h and g,
respectively, after upsampling by 2.
In DWT the original signal s is decomposed into
approximation and detail coefficients at the first stage;
while in the remaining stages, the decomposition is
IETE JOURNAL OF RESEARCH | Vol 55 | ISSUE 1 | JAN-FEB 2009

erformed on the approximation coefficients only, as

p
shown in Figure1, thus achieving the multi-resolution
analysis (MRA). This is called the Mallat algorithm or
Mallat-tree decomposition. Its significance lies in the
manner it connects the continuous-time multi-resolution to discrete-time filters. In this figure, the signal is
denoted by the sequence s(n), where n is an integer.
The low-pass filter is denoted by Go while the high-pass
filter is denoted by Ho. At each level, the high-pass filter
produces detailed information, cD; while the low-pass
filter associated with scaling function produces coarse
approximations, cA.
With the application of this approach, the time resolution
becomes arbitrarily good at high frequencies, while the
frequency resolution becomes arbitrarily good at low
frequencies. The filtering and decimation process is
continued until the desired level is reached.
After wavelet decomposition, the noise in a signal can
be further removed by wavelet de-noising. There are
two methods which can be adopted to remove the noise
using wavelet de-noising. The first method is forced
de-noising. This method turns selected high-frequency
coefficients to zero in wavelet decomposition structure.
After reconstruction, the results of this method are
smooth. But there is a chance of losing the useful highfrequency parts of the original signal. The second method
is threshold de-noising. This method gives the value of
threshold based on various thresholding algorithms
like universal threshold, rigorous SURE threshold and
minimax threshold [8]. The threshold-based de-noising
improves the de-noised results significantly. These
de-noised decomposition coefficients are then reconstructed using wavelet reconstruction.
Figure 2 shows the reconstruction of the original signal
from the wavelet coefficients. Basically, reconstruction
is the reverse process of decomposition. For reconstruction purposes, at each level, after upsampling,
s(n)

H0

G0

cD1

cA1

H0

G0

cD2

cA2

H0

cDi

G0

cAi

Figure 1: Multi-level wavelet decomposition tree.

'
'
'L

+
VQ

+
$

+
$

$L




*

*

*

Figure 2: Multi-level wavelet reconstruction tree.

11

[Downloaded free from http://www.jr.ietejournals.org on Sunday, July 04, 2010, IP: 123.238.23.177]
Chourasia VS and Mittra AK: A Comparative Analysis of De-noising Algorithm

the approximation coefficients are convolved with a

low-pass reconstruction filter G1 to obtain the reconstructed approximation Ai, while the detail coefficients
are convolved with a high-pass reconstruction filter H1
to obtain the reconstructed detail Di. The low- and highpass decomposition filters together with the low- and
high-pass reconstruction filters are called quadrature
mirror filters (QMFs). This process is continued through
the same number of levels as in the decomposition
process to obtain the original signal [9-12].

where n denotes the length of the signal and is the

standard deviation. This threshold was determined in
an optimal context for soft thresholding with random
Gaussian noise. This scheme is very easy to implement
but typically provides a threshold level larger than other
thresholding algorithms, therefore resulting in smoother
reconstructed data. Also, such estimation does not take
into account the content of the data but only depends
on the data size.

3.

The minimax threshold de-noising algorithm consists of

an optimal threshold that is derived from minimizing the
constant term in an upper bound of the risk involved in
the estimation of the signal. This threshold level depends
on the noise and signal relationships in the input data,
and it is given by

DE-NOISING ALGORITHMS

Three steps are required for de-noising of signals:

Decomposition, thresholding and reconstruction [13].
The first and last steps are performed with the selection of suitable wavelet family and mother wavelet.
The thresholding step is the selection of threshold
level for de-noising of the signals. The thresholding
algorithms commonly employed for de-noising of the
fPCG signals are
i) universal threshold
ii) minimax threshold
iii) rigorous SURE threshold
These de-noising algorithms can be divided into linear and
nonlinear methods. The linear method is independent of
the size of empirical wavelet coefficients, and therefore the
size of the coefficient by itself is not taken into account.
It assumes that signal noise can be found mainly in finescale coefficients and not in coarse scales. The nonlinear
method is based on the idea that the signal noise can be
found in every coefficient and is distributed over all scales.
It can be applied in two ways: Hard thresholding rule and
soft thresholding rule. In hard thresholding, if the value
of the coefficient is less than the defined value of threshold,
then the coefficient is scaled to zero; otherwise, the value
of the coefficient is maintained as it is. In soft thresholding, if the value of the coefficient is less than the defined
value of threshold, then the coefficient value is scaled to
zero; otherwise, the value of coefficient is reduced by the
amount of defined value of threshold [14].
3.1 Universal Threshold
The universal threshold de-noising algorithm is also known
as VisuShrink. It uses a fixed threshold form given by
= 2 log (n) ,

(6)

0LF

$XGLR
3UHDPSOLILHU

$FWLYH
/RZ3DVV
)LOWHU

3.2 Minimax Threshold

= n,

(7)

where n is determined by a minimax rule such that the

maximum risk of estimation error across all locations of
the data is minimized.
3.3 Rigorous SURE Threshold
The de-noising algorithms described previously use
global thresholds, that is, the computed threshold is
applied to all wavelet coefficients. The rigorous SURE
threshold algorithm describes a scheme that uses a
threshold value j at each resolution level j of the
wavelet coefficients. This algorithm is also known as
SureShrink, and it uses the Steins unbiased risk estimate
(SURE) criterion to get an unbiased estimate.

4.

COMPARISONS AND RESULTS

The fPCG signals are recorded from the maternal

abdominal surface using a highly sensitive and inexpensive data-recording module (DRM) [15]. The block
diagram of the DRM is shown in Figure 3.
The output of the sensor is fed to a separate pre-amplifier,
which provides high amplification and better noise rejection. These signals are then low-pass filtered with the help
of an active low-pass filter. A power amplifier further
strengthens the output signal from the filter. The recordings are obtained from pregnant women with gestation
age between 36 and 40 weeks. The signals are recorded

$XGLR3RZHU
$PSOLILHU

'DWD$FTXLVLWLRQ
+HDG
3KRQH

Figure 3: Block diagram of the data acquisition system.

12

IETE JOURNAL OF RESEARCH | Vol 55 | ISSUE 1 | JAN-FEB 2009

[Downloaded free from http://www.jr.ietejournals.org on Sunday, July 04, 2010, IP: 123.238.23.177]
Chourasia VS and Mittra AK: A Comparative Analysis of De-noising Algorithm

with a sampling frequency of 8000 Hz, 16-bit resolution,

and saved in a personal computer for further processing.
Figure4 shows the waveform of one of these fPCG signals.
All the three de-noising algorithms are simulated in
Matlab for de-noising of the fPCG signals. The meansquared error (MSE) is used to evaluate the performance
of the presented approach for the selection of appropriate de-noising algorithm. It can be obtained using the
following expression:
MSE =

in=1 (s se )i2 
n

(8)

where n denotes the length of the signal, s represents

the original signal and se is the estimated signal obtained
from the de-noised wavelet coefficients. Figure 5 is a test
signal generated by adding simulated random noise in
the original fPCG signal, which is shown in Figure 4.
This simulated random noise is analogous to the noise
produced because of maternal organ sounds, fetal
movement effects and ambient noise [16].
The wavelet analysis of this signal is performed with
five levels of decomposition using fourth-order Coiflets
wavelet. This selection of mother wavelet is random
and based on the fact that it possesses all the properties needed for analysis of the fPCG signals. All the
three algorithms with soft or hard thresholding rule are
applied for de-noising of the fPCG signal [17]. Figure 6
shows the waveforms of the fPCG signal obtained using
these optimal wavelet functions.
Finally, the efficacy of the method is evaluated using
fPCG signals (S1-S5) recorded from five different
subjects. All the de-noising algorithms are applied for
de-noising of these fPCG signals using selected mother
wavelets, and the respective results generated are
depicted in tabular form [Table 1] below.

Figure 4: Waveform of the fPCG signal.

IETE JOURNAL OF RESEARCH | Vol 55 | ISSUE 1 | JAN-FEB 2009

Table 1 shows a comparison among three de-noising

algorithms with soft or hard thresholding rule. The rigorous SURE threshold algorithm with soft thresholding
rule yields the best estimation with considerably smaller
MSE as compared to other algorithms for de-noising of
fPCG signals.

5.

CONCLUSION

Wavelet transforms have become a well-known useful

tool for various biomedical signal processing. This paper
presents an approach to select the appropriate de-noising
algorithm for de-noising of the fPCG signals. The performances of universal threshold, minimax threshold
and rigorous SURE threshold algorithms along with
soft and hard thresholding rules have been compared
for de-noising of the fPCG signal. The comparison is
based on evaluating the MSE of the original signal and
the estimated signal. The performance of the system is
validated using fPCG signals recorded from five different subjects. The results show that the rigorous SURE
Table 1: Comparison of de-noising algorithms
De-noising algorithm
Universal threshold
algorithm with soft
thresholding rule
Universal threshold
algorithm with hard
thresholding rule
Minimax threshold
algorithm with soft
thresholding rule
Minimax threshold
algorithm with hard
thresholding rule
Rigorous SURE
threshold algorithm with
soft thresholding rule
Rigorous SURE
threshold algorithm with
hard thresholding rule

S4

S5

0.5826 0.5744 1.1137

S1

S2

S3

0.2615

0.2817

0.5906 0.5666 1.1137

0.2615

0.2817

0.5371 0.5716 1.1137

0.2712

0.2817

0.8199 0.7972 1.1137

0.8162

0.8092

0.4661 0.5531

1.081

0.2539

0.2669

1.0091 0.8633

1.254

1.0062

0.8851

Figure 5: fPCG signal with additive noise.

13

[Downloaded free from http://www.jr.ietejournals.org on Sunday, July 04, 2010, IP: 123.238.23.177]
Chourasia VS and Mittra AK: A Comparative Analysis of De-noising Algorithm

Figure 6: De-noised fPCG signal using different algorithms: (a) sqtwolog(s), (b) sqtwolog(h), (c) minimaxi(s), (d) minimaxi(h),
(e) rigrsure(s), (f) rigrsure(h).

14

IETE JOURNAL OF RESEARCH | Vol 55 | ISSUE 1 | JAN-FEB 2009

[Downloaded free from http://www.jr.ietejournals.org on Sunday, July 04, 2010, IP: 123.238.23.177]
Chourasia VS and Mittra AK: A Comparative Analysis of De-noising Algorithm

threshold algorithm with a soft thresholding rule has

the best performance for de-noising of fPCG signals
when using fourth-order Coiflets wavelet. The proposed
approach is simple and proves to be effective when
applied for the selection of de-noising algorithm for the
fPCG signals. The main object is to improve the SNR of
the fPCG signals for accurate determination of FHR and
further diagnostic applications.

6.

6.

7.

8.

ACKNOWLEDGeMENT

The fetal heart sound recordings were taken at the District

Government Women Hospital. The authors of this paper thank
Dr. Shirish Ratnaparkhi (gynecologist) and Dr. (Mrs.)Megha
Ratnaparkhi (obstetrician) for their support in taking the
records with the help of developed prototype instrument.
The authors also thank the pregnant ladies who volunteered
to participate in the clinical test.

9.

10.
11.
12.

REFERENCES
1.
2.

3.

4.

5.

R.K. Freeman, and T.J. Garite, Fetal Heart Rate Monitoring, p. 7-17,
Williams and Wilkins, 1981.
P. Varady, L. Wildt, Z. Benyo, and A. Hein, An advanced method
in fetal phonocardiography, Computer Methods and Programs in
Biomedicine, vol. 71, pp. 283-96, 2003.
J. Chen, K. Phua, Y. Song, and L. Shue, A portable phonocardiographic
fetal heart rate monitor, Proceedings of the International Symposium
of the IEEE on Circuits and Systems, pp. 2141-4, 2006.
V. Padmanabhan, R. Fischer, J.L. Semmlow, and W. Welkowitz, High
sensitivity PCG transducer for extended frequency applications,
Proceedings of the Annual International Conference of the IEEE on
Engineering in Medicine and Biology Society, Images of the TwentyFirst Century, vol. 1, pp. 57-8, 1989.
F. Kovacs, Cs. Horvath, M. Torok, and G. Hosszu, Long-term
phonocardiographic fetal home monitoring for telemedicine
systems, 27th Annual International Conference of the IEEE on
Engineering in Medicine and Biology Society, pp. 3946-9, 2005.

13.
14.

15.

16.

17.

J. Nagel, New diagnostic and technical aspects of fetal

phonocardiography, European Journal of Obstetrics, Gynecology
and Reproductive Biology, vol. 23, pp. 295-303, 1986.
G. Vasios, A. Prentza, D. Blana, E. Salamalekis, P. Thomopoulos,
D. Giannaris, and D. Koutsouris, Classification of fetal heart rate
tracings based on wavelet-transform and self-organizing-map neural
networks, Proceedings of the 23rd Annual International Conference
of the IEEE on Engineering in Medicine and Biology Society, vol. 2,
pp. 1633-6, 2001.
X. Yang, P. Li, Z. Xin, Z. Bian, and B. Wang, De-Noising of the doppler
fetal heart rate signal with wavelet threshold filtering based on
spatial correlation, The 1st International Conference of the IEEE on
Bioinformatics and Biomedical Engineering, ICBBE 2007, pp. 928-31,
2007.
M.R. Raghuveer, and A.S. Bopardikar, Wavelet Transforms:
Introduction to Theory and Applications, pp. 25-50, Pearson
Education Pvt. Ltd., Singapore, Indian Branch, 2002.
I. Daubechies, Ten Lectures on Wavelets, Society for Industrial and
Applied Mathematics New York, 1992.
S.G. Mallat, A Wavelet Tour of Signal Processing, Academic Press,
1999.
S.G. Mallat, A theory of multiresolution signal decomposition: The
wavelet rpresentation, IEEE Transactions on Pattern Analysis and
Machine Intelligence, vol. 11, pp. 674-93, 1989.
Matlab Wavelet Toolbox Users Guide, Mathwork Incorporation;
www.mathworks.com.
M.C.E. Rosas-Orea, M. Hernandez-Diaz, V. Alarcon-Aquino, and
L.G. Guerrero-Ojeda, A comparative simulation study of wavelet
based denoising algorithm, Proceedings of the 15th International
Conference of the IEEE on Electronics, Communications and
Computers, pp. 125-130, 2005.
V.S. Chourasia, and A.K. Mittra, Development of data acquisition
module for a non-invasive fetal monitoring system, International
Journal of Biomedical Signal Processing, in press, 2008.
A.K. Mittra, N.K. Choudhary, and A.S. Zadgaonkar, Development of
an artificial womb for acoustical simulation of mothers abdomen,
International Journal of Biomedical Engineering and Technology,
vol.1, no. 3, pp. 315-28, 2008.
W.G. Morsi, and M.E. El-Hawary, The most suitable mother wavelet
for steady-state power system distorted waveforms, Canadian
Conference of the IEEE on Electrical and Computer Engineering,
pp.17-22, 2008.

AUTHORS
Vijay Chourasia graduated in Electronics Engineering
from RTM Nagpur University, Nagpur. He is pursuing
his Master of Engineering (By Research) from RTM
Nagpur University, Nagpur. He is Senior Lecturer in the
Department of Electronics Engineering in Manoharbhai
Patel Institute of Engineering & Technology, Gondia,
India. He has more than 15 years experience in the field of academics and
has about 10 research publications in various national and international
conferences and journals. His area of research is Biomedical Instrumentation,
Biomedical Signal Processing and Soft Computing.
E-mail: chourasiav@gmail.com

A. K. Mittra graduated in Electrical Engineering

from GGD University Bilaspur and Post Graduated
in Electronics and Control Engineering from BITS
Pilani. He is pursuing his Doctoral Research from RTM
Nagpur University, Nagpur. Currently he is an Assistant
Professor in the Department of Electronics Engineering
in Manoharbhai Patel Institute of Engineering &
Technology, Gondia, India. He has a vast experience in academic field and he
has more than 25 research publications in various national and international
conferences and journals. His area of research is Biomedical Instrumentation,
Biomedical Signal Processing and Soft Computing.
E-mail: akmittra@gmail.com

DOI: 10.4103/0377-2063.51322; Paper No JR 85_08; Copyright 2009 by the IETE

IETE JOURNAL OF RESEARCH | Vol 55 | ISSUE 1 | JAN-FEB 2009

15