International Journal of Computer Applications (0975 – 8887)
Volume 23– No.7, June 2011
1
Removal of 50Hz PLI using Discrete Wavelet Transform
for Quality Diagnosis of Biomedical ECG Signal
Ramesh D. Mali
M.Tech. Student, Department of
Electronics Tech, Shivaji
University, Kolhapur (MS) India
Mahesh S. Khadtare
Research Student, Dept. of
Electronics, I Square IT,
Hinjawadi, Pune, India
Dr. U.L Bombale
Department of Electronics Tech.
Shivaji University, Kolhapur
(MS) India
ABSTRACT
Electrocardiogram (EKG or ECG) is an important electrical
activity of the human Heart. ECG is used for the primary
diagnosis of heart diseases since it shows the
electrophysiology of the heart and the ischemic changes
that may occur like the myocardial infarction, conduction
defects, and arrhythmia. But, in real condition, ECG is
often corrupted by different artifacts and noises. For the
purpose of quality diagnosis, the ECG signal must be
clearly de-noised to remove all noises and artifacts from the
signal. In this paper, we present the Wavelet Transform, a
new approach in digital signal processing to filter the ECG
signal. Different ECG signals from MIT/BIH arrhythmia
database are used with added 10dB, 5dB & 0dB Power
Line Interference (PLI) noise which is common in ECG
signal. The results were evaluated using M ATLAB
software. Basically, two synthesis parameters M ean Square
Error (M SE) and Signal to Noise ratio (SNR) have been
used. The prime aim of this paper is to adapt the discrete
wavelet transform (DWT) to improve the (ECG) signal
quality for better clinical diagnosis. The evaluated results
have been compared with Butterworth IIR filter. The
proposed method shows improvement in output SNRo for
5dB noise is 98.5% and for 10dB noise is 95.7%.
Keywords
Wavelet de-noising, ECG Signal and Noise, Discrete
Wavelet Transform, Thresholding
1. INTRODUCTION
The electrical activity of Heart are represented by the
Electrocardiogram (ECG) is easily susceptible to the
various kinds o noise : the Electromygram (EMG) signal,
Baseline Wander signal, 50/60 Hz Power Line Interference
(PLI) etc. However, Noise contamination to these noises
can degrade the ECG signal and cause to loss of clinical
information. Thus, the filtering of ECG is necessary to
conserve the useful information and to remove such noises.
This work remains as a challenge. The American Heart
Association (AHA) has defined standard filtering
requirements for clinical ECG equipment [1]. Generally,
adequate ECG de-noising algorithms and procedures
should have properties: Improved signal to noise ratio
(SNR) and Mean Square Error (MSE) for obtaining clean
and readily observable recordings, yielding the subsequent
use of straightforward approaches for automatic detection
of characteristic points in the ECG signal and
reorganization of its specific waves and complexes.
Many researchers have worked on the ECG signal de-
noising. Power line interference (PLI) and Baseline wander
in the ECG signal are the major problem in the diagnostic
of ECG. Different researchers have worked on removal of
AC interference (50/60 Hz) so as to retain basic ECG signal
characteristics.
G. Umamaheswara Reddy et.al.[2] proposed a new
thresholding technique compromise between hard and soft
thresholding for ECG signal Denoising using evaluation
criteria: Mean Square Error (MSE) and output SNR.
Donoho and Johnstone [3] proposed a very simple
thresholding procedure based on the Discrete Wavelet
Transform with universal threshold for getting better signal
from noisy data is very much suitable for Non-stationary
ECG signal.
ECG signals are very low voltage amplitude (1mV)
signals and easily susceptible to many noises and artifacts
and one of them is Power Line Interference Noise. This
noise causes the problem in analysis of low voltage level
signals like ECG. Manpreet Kaur, Birmohan Singh [4]
proposed a combination of Notch and Moving Average
method for PLI reduction. M ahesh S. Chavan, R.A.
Aggarwala, M .D.Uplane [5] has used Digital FIR Filters
based on Rectangular window for the power line noise
reduction. Removal of 60Hz PLI and ECG signal
amplification of Remote ECG systems was developed by
Ying-Wen Bai et.al. [6]. A novel method for elimination of
PLI and BW in ECG signal was developed by Zhi-Dong
Zhao et.al.[7]. Baseline wanders and power line
interferences removing are the first step in quality diagnosis
biomedical ECG signal [8-13]. M ikhled Alfaouri and
Khaled Daqrouq [14] have proposed the combination
technique to remove power line interference.
2. METHODS
The ECG signal is easily corrupted by noises such as
Gaussian noise, baseline wander, EMG, Power line
interference (PLI) and so on.
The method can be divided in into the following steps:
2.1 Noise Generation and Addition:
The 50/60 Hz Power line interference noise is generated
and added into the original ECG signal samples taken from
International Journal of Computer Applications (0975 – 8887)
Volume 23– No.7, June 2011
2
the MIT/BIH database. The process of adding noise to
original signal is mathematically shown as:
F (n) = X (n) + D (n), n = 1, 2, 3… N.
Where, X (n) is the original ECG signal,
D (n) is the 50/60 Hz PLI noise,
F (n) is the Noisy ECG signal.
2.2 Basic Steps:
The basic blocks utilized in the proposed system
are shown in Fig 1.
F(n)
Fig 1: Proposed system blocks
In transform domain, we perform DWT of the
signal. Second we pass the transform through a threshold to
remove the coefficients below a certain value. In inverse
transform domain, we take the Inverse DWT (IDWT) to
reconstruct the original ECG signal. Thresholding or
shrinking the wavelet transform will remove the low
amplitude noise or undesired signals and any noise overlap
as little as possible in the frequency domain and linear
time-invariant filtering will approximately spare them. It is
the localizing or concentrating properties of the wavelet
transform that make it particularly effective when used with
this nonlinear method.
2.3 Wavelet Transform:
A wavelet is simply a small wave which has
energy concentrated in time to give a tool for the analysis
of transient, non-stationary or time-varying phenomena
such a wave shown in figure 2.
Fig 2: Wavelet function
Wavelet transform is an emerging tool for the de-
noising of non-stationary signals like ECG. There are
number of wavelet families like Haar, Daubechies (Db),
Symlet etc for analysis and synthesis of signal. Proper
selection of wavelet basis function is plays an vital in de-
noising. Since Db is mostly morphologically similar to the
ECG signal, so in present work Db is used in de-noising
and its comparative results with IIR filter and HAAR
wavelet are discussed. In Discrete Wavelet transform
(DWT), original signal is decomposed and reconstructed
using the low pass h(n) and high pass g(n) filter bank
tree as shown in Fig 3.
Fig 3: Filter bank tree a) Decomposition (DWT)
b) Reconstruction (IDWT)
2.3.1 Thresholding Method
In discrete wavelet transform, threshold is
applied to the signal after passing through the DWT, and
then IDWT is taken.
(1)
Where T is the threshold, N is no. of samples,
, is the standard deviation of noise.
Two thresholding methods are used namely Hard threshold
and Soft threshold.
2.4 Evaluation Criteria:
2.4.1 Estimation of Mean Square Error (MSE):
The M SE value is estimated between the de-noised ECG
signal and original ECG signal taken from MIT/BIH
database is given by eq. (2).
(2)
Where N is the length of ECG signal,
is the original
ECG signal and is the de-noised ECG signal.
2.4.2 Estimation of Signal to Noise ratio (SNR):
Transform
domain
Threshold
Inve rse Trans
Domain
International Journal of Computer Applications (0975 – 8887)
Volume 23– No.7, June 2011
3
The output SNR is given by eq. (3).
(3)
SNRo values to determine the wavelet function for
Denoising ECG signal.
2.5 Butterworth IIR Notch Filte r:
The Butterworth IIR Notch filter is designed
using the MATLAB FDATool and it is required to specify
the sampling frequency Fs, Filter order and cut-off
frequency (Fc1 and Fc2).
3. RESULTS
In this section, we discussed on the result obtained with the
experimental work done. In the proposed de-noising
algorithm, the five set of ECG records of M IT/BIH
database were used and sampling frequency is set to 360Hz
and added with 50 Hz Power line Interference noise with
different input SNR values. The effectiveness of proposed
algorithm was determined by the MSE and output SNRo
values. The IIR notch filter, Haar wavelet transform and
Daubechies wavelet transform filters were used in proposed
algorithm to obtain quality de-noised ECG signal for
diagnosis and analysis. The obtained results were discussed
in below sub-section.
3.1 Simulation Study
We explore our proposed algorithm result done in
MATLAB®7.1 simulations to 0dB and 5 dB noisy ECG
data segments. The Fig 4 shows the simulated results of IIR
notch filter with 5dB noisy ECG signal but it unable to
minimize the ringing effect seen. The Table 1 shows the
MSE and SNRo values for input SNR values 0dB and 5dB
noisy ECG signal. These average results clearly shows that
as noise level goes on increasing from 5dB to 10db, MSE
increases and SNRo decreases. When input SNR increases
from 5dB to 10dB, we found 49.1% output SNRo values.
Fig 4: De-noising of ECG signal using IIR Notch Filter.
Table 1: The MS E and S NR values for the de-noising
algorithm using IIR Notch filter:
EC G
Data
I/P SNR = 5 dB
I/P SNR =10dB
MSE
SNR
MSE
SNR
Sample1
0.001747
15.059091
0.013322
6.235842
Sample2
0.003732
15.480584
0.018024
8.641819
Sample3
0.000659
15.242976
0.007164
4.880675
Sample4
0.004091
13.006576
0.011408
8.552229
Sample5
0.00156
15.784318
0.012796
6.64485
Sample6
0.003327
13.965597
0.01149
8.582474
Ave rage
0.0025193
14.7565237
0.0123673
7.2563148
Next step in our work, we apply the Haar wavelet
transform on the 5dB and 10dB noisy ECG signal. Fig 5
shows the simulated result of the same algorithm and Table
2 shows the MSE and SNRo values for the same transform.
Obtained result clearly shows that when input SNR values
increases from 5dB to 10dB, we found 65.1% output
SNRo. Also, we observe there is no ringing effect as seen
in IIR notch filter shown in Fig 4.
International Journal of Computer Applications (0975 – 8887)
Volume 23– No.7, June 2011
4
Fig 5: De-noising of ECG signal using Haar wavelet
transform.
Table 2: The MS E and S NR values for the de-noising
algorithm using Haar wavelet transform:
EC G
Data
I/P SNR = 5 dB
I/P SNR =10dB
MSE
SNR
MSE
SNR
Sample1
0.007364
8.810233
0.016133
5.40428
Sample2
0.010199
11.114828
0.022828
7.615621
Sample3
0.004035
7.373314
0.009739
3.547186
Sample4
0.015903
7.109659
0.023051
5.497628
Sample5
0.005688
10.166009
0.013983
6.259636
Sample6
0.007675
10.334941
0.015158
7.379248
Ave rage
0.0084773
9.1514973
0.0168153
5.9505998
In the very next stage, we apply the Daubechies
wavelet transform (Db2, Db3 and Db4) to obtain the noise
free ECG signal. Fig 6, Fig 7, Fig 8 shows the simulated
result of the Daubechies algorithm (Db2, Db3 and Db4).
The Daubechies wavelet transforms results shows less
distortion in original signal. Table 3, Table 4, Table 5
shows the M SE and SNR values for the Daubechies (Db2,
Db3 and Db4) wavelet transform. Table 2 indicates that
average SNRo values for 5dB is 10.84692 while for 10dB,
it is 9.03848183. In Table 3, the Db3 transform shows that
output SNRo of 12.14687 for 5dB and 11.6358998 for
10dB. Also Fig 7 shows the 98.5% de-noised ECG signal.
Fig 6: De-noising of ECG signal using Daubechies (Db2)
wavelet transform.
Fig 7: De-noising of ECG signal using Daubechies (Db3)
wavelet transform.
Fig 8: De-noising of ECG signal using Daubechies (Db4)
wavelet transform
International Journal of Computer Applications (0975 – 8887)
Volume 23– No.7, June 2011
5
Table 3: The MS E and S NR values for the de-noising
algorithm using Daubechies (Db2) wavelet transform.
EC G
Data
I/P SNR = 5 dB
I/P SNR =10dB
MSE
SNR
MSE
SNR
Sample1
0.006164
9.583113
0.008672
8.100461
Sample2
0.004958
14.24753
0.008706
11.801883
Sample3
0.003126
8.482568
0.004959
6.47832
Sample4
0.012636
8.108472
0.014794
7.423701
Sample5
0.004296
11.38538
0.006858
9.353469
Sample6
0.003901
13.27448
0.006476
11.073057
Ave rage
0.00584683
10.84692
0.008411
9.03848183
Table 4: The MS E and S NR values for the de-noising
algorithm using Daubechies (Db3) wavelet transform.
EC G
Data
I/P SNR = 5 dB
I/P SNR =10dB
MSE
SNR
MSE
SNR
Sample1
0.005685
9.934195
0.006016
9.687997
Sample2
0.005718
13.62801
0.006198
13.278012
Sample3
0.002995
8.668531
0.003244
8.320986
Sample4
0.009267
9.454963
0.009593
9.305106
Sample5
0.003197
12.6684
0.003593
12.161591
Sample6
0.001164
18.5271
0.001631
17.061707
Ave rage
0.004671
12.14687
0.005046
11.6358998
Table 5: The MS E and S NR values for the de-noising
algorithm using Daubechies (Db4) wavelet transform.
EC G
Data
I/P SNR = 5 dB
I/P SNR =10dB
MSE
SNR
MSE
SNR
Sample1
0.005017
10.47674
0.006338
9.462156
Sample2
0.004271
14.89539
0.006111
13.339384
Sample3
0.0033
8.247336
0.004181
7.219191
Sample4
0.009445
9.37242
0.010577
8.880813
Sample5
0.004029
11.66326
0.005327
10.450469
Sample6
0.002339
15.49592
0.003661
13.549909
Ave rage
0.0047335
11.69185
0.006033
9.8704026
Table 6 shows the experimental result obtained when the
input signal SNR is increased from 0 dB to 5 dB, 5dB to 10
dB and also 0 dB to 10 dB. In this table, the calculated
average value of output SNRo for the different ECG signals
from the MIT/BIH database have been shown. The results
clearly show that for IIR notch filter output SNRo varies
from 51.9% to 25.53%. Similar to this, for Haar wavelet
transform, output SNRo varies from 89.12% to 57.95%.
This shows that Haar wavelet transform is better than IIR
notch filter under noisy condition but it affects the shape of
the signal and disturbs the wave. This problem is fixed with
Daubechies wavelet transform Db3 found very good result
and its output SNRo varies from 98.50% to 94.30%.
Table 6: The output SNRo values for different methods
used for different input S NR values.
Method
input noise in dB
0-5 dB
5-10 dB
0 - 10 dB
IIR
51.90%
49.10%
25.53%
Haar
89.12%
65.10%
57.95%
Db2
95.22%
83.30%
79.34%
Db3
98.50%
95.70%
94.30%
Db4
96.60%
84.40%
81.50%
4. CONCLUSION
The proposed work illustrates the effect of the wavelet
thresholding on the quality reconstruction of ECG signal.
The IIR notch filter applied directly to the non-stationary
signal like ECG has shown more ringing effect. Daubechies
Db3 wavelet transform is the best method to de-noise the
noisy ECG signals. For 5dB and 10dB input noise value,
Db3 wavelet transform shows the output SNRo value
98.50% and 95.7% respectively with respect to other IIR,
Haar, Db2 and Db4 to 95.70% which is very good for de-
noising signal. We conclude that our work shows
Daubechies wavelet transform performs the better than
other methods. We can further illustrate this work by
including other wavelets family to estimate the better
quality de-noising of ECG signal.
5. REFFERENCES
[1] Kligfield P, Gettes L, Bailey J, et al.
“Recommendations for the Standardization and
Interpretation of the Electrocardiogram” , J Am Coll
Cardiol 2007;49:276–281.
[2] G. Umamaheswara Reddy, Prof. M . M uralidhar, Dr.
S. Varadarajan,” ECG De- Noising using improved
thresholding based on Wavelet transforms”, IJCSNS
International Journal of Computer Science and
Network Security, VOL.9 No.9, September 2009.
[3] David L. Donoho, “De-noising by soft thresholding”,
IEEE Trans. Trans. Inform. Theory, vol. 41, pp. 613-
627, 1995
[4] M anpreet Kaur, Birmohan Singh,” Powerline
Interference Reduction in ECG Using Combination