A novel adaptive algorithm for removal of power line interference from ecg signal

Sci.Int.(Lahore),28(1),139-143,2016 ISSN 1013-5316; CODEN: SINTE 8 139
Jan.-Feb.
A NOVEL ADAPTIVE ALGORITHM FOR REMOVAL OF POWER LINE
INTERFERENCE FROM ECG SIGNAL
Sadaf Khan1
, Syed Muhammad Anwar1
, Waseem Abbas2
, Rizwan Qureshi3
1
Department of Computer Engineering, University of Engineering & Technology, Taxila, Pakistan,
2
Department of Computer Sciences, COMSATS Institute of Information Technology, Abbottabad, Pakistan
3
Department of Electrical Engineering, Institute of Space Technology, Islamabad, Pakistan
[1
sadafgondal@yahoo.com, 2
s.anwar@uettaxila.edu.pk, 3
waseem_abbas@live.com, 4
ritz_gemini12@yahoo.com]
ABSTRACT: During data acquisition and transmission of biomedical signals like electrocardiography (ECG), different
types of artifacts are embedded in the signal. Since an ECG is a low amplitude signal these artifacts greatly degrade the
signal quality and the signal becomes noisy. The sources of artifacts are power line interference (PLI), high frequency
interference electromyography (EMG) and base line wanders (BLW). Different digital filters are used in order to reduce
these artifacts. ECG signal is a non-stationary signal, it is difficult to find fixed filters for the removal of interference
from the ECG signal. In order to overcome these problems adaptive filters are used as they are well suited for the non-
stationary environment. In this paper a new algorithm “Modified Normalized Least Mean Square” has been proposed. A
comparison is made among the new algorithm and the existing algorithms like LMS, NLMS, Sign data LMS and Log
LMS in terms of SNR, convergence rate and time complexity. It has been observed that the performance of new algorithm
is superior to the existing ones in terms of SNR and convergence rate however it is more complex than the other
algorithms. Results of simulations in MATLAB are presented and a critical analysis is made on the basis of convergence
rate, signal to noise ratio (SNR), and computational time among the filtering techniques.
Keywords: Electrocardiography, Power Line Interference, Least Mean Square, Normalized Least Mean Square, Signal to Noise Ratio
1. INTRODUCTION
Electrocardiogram (ECG) is a very significant
physiological signal of the human body. It contains
information about the performance of the human heart. A
doctor can detect many types of irregularities /diseases by
analyzing waveform of the ECG signal. In short ECG to a
human body is the same as a trouble shoot message to a
digital system since a trouble shoot message detects
hardware malfunction. Similarly ECG waveform can
trigger alarms in case of danger or prior heart attacks.
One of the major issues in biomedical signal processing is
the presence of outliers/artifacts. The major artifacts arise
in ECG signal may be due to technical fluctuation like
Power line interference, Electromagnetic fields of different
electronic devices, Instrumentation noise or it may have
biological resources such as random body movements,
muscular reaction and respiratory movements[8].These
artifacts affect the signal quality and lead to wrong clinical
diagnosis. The artifacts need to be suppressed or removed
in order to achieve reliable and improved signal quality
and thus providing a higher level of health caring activity
to the subject. One of the major artifacts in ECG signal is
the Power line interference (PLI).The PLI is 50/60 Hz
depending upon the supply. The frequency content of ECG
signal is also nearly the same. Many techniques exists in
the literature to remove the PLI such as low pass filter,
Notch filter [1]. Since these are fixed filters and most of
the information in the ECG signal has the same frequency
as PLI. So these filters eliminate some useful information
from the ECG signal. Adaptive filter are a useful technique
for the removal of artifacts from biomedical signal. In
2010, Dr K.L Yadav and S.Singh have make use of LMS
and RLS adaptive algorithm in the paper titled
“Performance evaluation of different adaptive filter for
ECG signal processing”[2]. In another paper “Efficient
sign based normalized adaptive filtering techniques for
cancellation of artifact in ECG signal; Application to
wireless biotelemetry” by M.Z Rahman, R.A Shaik and
D.V Rami Koti Reddy have make use of various simple
and effective sign based normalized adaptive filters for
cancellation of artifacts in ECG ,which are found
computationally superior and efficient having multiplier
free for update loop[3].For the removal of artifacts from
ECG signal an unbiased and normalized adaptive filtering
techniques were also used in the paper title “Filtering of
Noise in Electrocardiographic Signals Using An Unbiased
and Normalized Adaptive Artifact Cancellation System” by
Yunfeng Wu, Rangaraj M. Rangayyan Ye Wu, and Sin-
Chun Ng [9] In 2012, performance comparison is made
between Modified LMS and RLS algorithm in the paper
title “Performance comparison of modified LMS and RLS
algorithms in Denoising of ECG Signal”[4]. An ECG
signal consists of sudden large peaks. In existing adaptive
algorithms steady state is destroyed when a sudden large
peak comes and hence the primary convergence is
destroyed. We have proposed a novel approach to hold the
primary convergence. Our proposed approach produces
better results than the existing ones and hence can be
applied to automatic ECG analysis. In E-health or
automatic ECG analysis PLI should be more accurately
filtered because it overlaps the frequency spectrum of the
ECG signal which may lead to wrong ECG analysis or
diagnosis. For removal of PLI, Our proposed approach
produces better results in term of convergence rate and
SNR. So it can be applied in automatic ECG analysis
applications.
2. Adaptive Algorithm
FIR based adaptive filters are preferred due to their
stability. In adaptive noise cancellation system a refrence
signal i-e corelated with the noise is fed input to the
adaptive filter as shown in fig 1.Desired signal d(n) is the
combination of the primary signal that is informtion signal
x(n) and noise .

140 ISSN 1013-5316; CODEN: SINTE 8 Sci.Int.(Lahore),28(1),139-143,2016
Jan.-Feb.
d(n) = x(n) + noise (1)
y(n) is the filter output and is calculated using Equation
(2).
y(n)=wT
(n) * x(n) (2)
Filter output vector y(n) can be expressed by Equation (3)
y(n)=[y(0)y(1)y(2)………y(n-1)]T
(3)
x(n) and wT
(n) are input signal and tapped weight vector
and are expressed in vector form by Equation 4 and
Equation 5 respectively.
x(n)=[x(0)x(1)x(2)……x(n-1)] (4)
w(n)=[w(0)w(1)w(2)……w(n-1)]T
(5)
Error signal is obtained by subtracting the filter output y(n)
with the received signal d(n) and given by the Equation (6).
e(n)=d(n) –y(n) (6)
Figure 1. Adaptive Noise Cancellation System
2.1 Least Mean Square Algorithm
The LMS algorithm was first proposed by Widrow and
Hoff in 1959 through their studies of pattern recognition
[5]. It is based upon stochastic gradient algorithm which
finds optimal weights using the gradient vector to converge
on an estimated optimal solution. It is simple, less complex
and probably the most widely used algorithm.
The weight vectors in each iteration are updated in the
LMS algorithm according to the following equation.
w(n+1)=w(n) +µ * e(n) *x(n) (7)
Here w(n+1) is the adapted weight and w(n) is the
previous weight. µ is the step size that determines how
much of the error to be adjusted. It directly affects the
convergence rate. A large step size converges faster but
may not lead to the optimal solution whereas optimal
solution is achieved by selecting the smaller step size but it
converges slowly. e(n) is the error signal and x(n) is the
input signal.
2.2 Normalized Least Mean Square Algorithm (NLMS)
In simple LMS algorithm it is difficult to choose the step size.
This problem is solved by the NLMS algorithm. In the NLMS
algorithm step size becomes adaptive. Due to the adaptivity of
step size this algorithm found stable and have fast
convergence rate as compared to the LMS algorithm [10].
The coefficient weight vector update relation for the
NLMS algorithm is given by the following equation.
w(n+1)=w(n) +µ/[x(n)*x’
(n)] *e(n) *x(n) (8)
Here the step size is normalized using the above relation.
2.3 Sign Data LMS Algorithm (SDLMS)
The Sign Data Least Mean Square algorithm is obtained by
the well-known LMS algorithm by replacing the input signal
vector x(n) with the vector sgn{x(n)} [8].The advantage of
using the Sign Data LMS algorithm is the less computational
complexity as sign {.} of every function is calculated very
fast and it can be used in high speed bio telemetry
applications. The weight updating formula for the Sign Data
LMS algorithm is same as the simple LMS except that the
input signal is quantized by using the well-known sign
function.
w(n+1)=w(n) +µ * e(n) * sgn{x(n)} (9)
𝑠𝑔𝑛(𝑥) = {
1 𝑥 > 0
−1 𝑥 < 0
0 𝑥 = 0
The computational complexity of the Sign Data LMS
algorithm is also less than the simple LMS because of the
multiplication operation is replaced with the shifting
operation.
2.4 Log Least Mean Square (LLMS) Algorithm
In Log LMS algorithm, instead of quantizing the input
signal the error signal is quantized using the log function.
w(n+1)=w(n) +µ *Q{ e(n) }*x(n) (10)
Q (·) is the logarithmic quantizing function. It will simply
convert input into the power-of-two value [4] .The
quantizing function is defined by the following relation.
Q(z)=2Ɩ log
2
( Ɩz Ɩ)
* sgn(z) (11)
2.5 Proposed Modified Normalized Least Mean Square
(MNLMS) Algorithm
In above discussed algorithms the output of the system e(n) is
used to update the filter coefficients. Steady state is destroyed
when large peak come into the periodic life signal. It results
in signal distortion. The iteration function for the modified
NLMS is given by the following formula:
w(n+1)=w(n) +µ/[x(n)*x’
(n)] *Q{e(n)}*x(n) (12)
Where Q[.] represents the logarithmic quantization
function and is defined by equation 13.
Q(z)=α *2θ(z)
* sgn(z) (13)
In this algorithm we have divided the processing of ECG
signal in two different stages, Convergence stage and
Extracting stage. In first stage, the proposed algorithm
iterates with the rapid convergence rate to achieve a
primary convergence. When primary convergence is
achieved the algorithm enters into the extracting stage.
Θ(z)= log2 Ɩz/αƖ
At convergence stage α represents the power of two values
and is always less than 1.The term z/α is used to amplify
the ECG signal to integer type by shifting operation. When
convergence state is achieved primarily the algorithm
iterates for the extraction of information from the signal
using the following relation.

Jan.-Feb.
Θ(z)= - log2 Ɩz/αƖ
At extracting stage Switching between the above two
stages, a threshold value € is introduced. When output of
the system becomes less than or equal to € for large
number of consecutive iterations that is
|e(n)- e(n-1)| < €
Than we can conclude that the convergence state is
achieved and algorithm enters in the second stage
3. Implementation of Algorithms
Adaptive algorithms discussed in this paper are implemented
in MATLAB R2012b. Five records of the original biomedical
ECG signal were extracted from the record MIT-DB
Arrhythmia database. However the plots are shown only for
the record number 100 of 10 seconds recording of ECG signal
containing 3600 samples as the sampling rate of the signal is
360.The Power line interference (PLI) was generated in
MATLAB by using a sine wave of 60 HZ and is then added
to the original signal to make it noisy. The step size for linear
and nonlinear algorithms was chosen to be 0.005. The
reference signal was generated in MATLAB by using the
knowledge about the frequency content of the artifact. The all
discussed adaptive algorithms were implemented on this
noisy ECG signal and their performance was evaluated in
terms of SNR, Convergence rate and Time complexity.
4. SIMULATION RESULTS
In simulation input to adaptive filter is a signal correlated
with the PLI type noise. The filter order was chosen to be two
and step size chosen to be 0.005. Figure 2 & 3 shows the
original ECG signal and ECG signal corrupted by PLI type
noise.
Fig.ure 2. Original ECG signal
Figure 3. Corrupted ECG signal by PLI type noise
4.1 RESULTS OF REMOVING ARTIFACT FROM ECG
SIGNAL
Step size for each algorithm was taken to be 0.005.
Figure 4. Filtered signal & MSE plot using LMS
0 500 1000 1500 2000 2500 3000 3500 4000
1.7
1.8
1.9
2
2.1
2.2
2.3
2.4
2.5
ECG signal
Samples
Amplitude
0 500 1000 1500 2000 2500 3000 3500 4000
0.5
1
1.5
2
2.5
3
3.5
Samples
Amplitude
Corrupted ECG
0 500 1000 1500 2000 2500 3000 3500 4000
1.5
2
2.5
Filtered signal using LMS
Samples
Amplitude
0 500 1000 1500 2000 2500 3000 3500 4000
0
0.5
1
Mean Square Error
Samples
Amplitude

142 ISSN 1013-5316; CODEN: SINTE 8 Sci.Int.(Lahore),28(1),139-143,2016
Jan.-Feb.
Figure 5. Filtered signal & MSE plot using NLMS
Figure 6. Filtered signal & MSE plot using
SDLMS
Fig.ure 7. Filtered signal & MSE plot using Log LMS
Fig.ure 8. Filtered signal & MSE plot using proposed modified
NLMS
Table 1. Analysis of the algorithm in terms of SNR,
Convergence Rate & Time complexity.
0 500 1000 1500 2000 2500 3000 3500 4000
1.5
2
2.5
Filtered Signal using NLMS
Samples
Amplitude
0 500 1000 1500 2000 2500 3000 3500 4000
0
0.5
1
Mean Square Error
Samples
Amplitude
0 500 1000 1500 2000 2500 3000 3500 4000
1.5
2
2.5
Filtered signal using SDLMS
Samples
Amplitude
0 500 1000 1500 2000 2500 3000 3500 4000
0
0.5
1
Mean Square Error
Samples
Amplitude
0 500 1000 1500 2000 2500 3000 3500 4000
1.5
2
2.5
Filtered signal using LogLMS
Samples
Amplitude
0 500 1000 1500 2000 2500 3000 3500 4000
0
0.5
1
Mean Square Error
SamplesAmplitude
0 500 1000 1500 2000 2500 3000 3500 4000
0
1
2
3
Filtered signal using modified NLMS
Samples
Amplitude
0 500 1000 1500 2000 2500 3000 3500 4000
0
0.5
1
Mean Square Error
Samples
Amplitude
ECG Noise LMS NLMS SDLMS LLMS
MNL
MS
SNR PLI 21.67 19.04 19.76 21.86 22.17
Convergence 3000 1000 750 800 500
Time
complexity
(secs)
0.148 0.170 0.114 1.136 1.506

Jan.-Feb.
Fig.ure 9. Histogram of PLI type noise
5. FUTURE WORK AND DISCUSSION
We have analyzed the response of different adaptive
algorithms in this paper for the removal of PLI type noise
and we found that for PLI type noise the MNLMS gave
the best SNR and convergence speed at the cost of more
time complexity. Since an ECG signal consists of many
artifacts so in future one may design a multistage filter
having multiple reference inputs whose one filter should
be MNLMS type. ECG stills remain an open &
extensive area for research for signal processing .Wavelet
transform can be used for the extraction of ECG signal
and distance between two consecutive R-R peaks can be
calculated to determine the heartbeat. In automatic ECG
analysis the decision has to be made by the machine so if
the machine get more purified signal it can analyze the
ECG signal in a better way and the chances of error are
minimized. Furthermore, our proposed approach can be
implemented in e-health system for the better analysis of
biomedical signal.
REFERENCES
[1] Anna Esposito, Pierluigi D. Andria, “An Adaptive
Learning Algorithm for ECG Noise” WIRN VIETRI
2003, LNCS 2859, pp. 139–147, 2003.
[2] Dr. K. L. Yadav and Sachin Singh, “ Performance
evaluation of different adaptive filters for ECG signal
processing”, International Journal On Computer
Science and Engineering, vol. vol. 40, no. 5, pp. 1880–
1883, 2010
[3] Rafi Ahamed Shaik and D.V.RamaKoti Reddy ,
Muhammad ZiaUrRahman, "Efficient sign based
normalized adaptive filtering techniques for
cancelation of artifacts in ECG signals: Application to
wireless biotelemetry”, Journal of signal processing,
vol. vol. 91, no. 2, ,pp. 225–239, February 2011.
[4] Md.Zameari Islam, G.M Sabil Sajjad, Md.Hamidar
Rahman, Ajoy Kumar Dey, Md Abdul Matin Biswas,
A.K.M.J Hoque, „Performance Comparison of
Modified LMS and RLS Algorithms in Denoising of
ECG Signal”, International Journal of Engineering &
technology ,Volume 2 No. 3, March, 2012.
[5] R. Ranjit Kumar Syed Attequr Rehman, "Performance
Comparision of Different Adaptive Filters Algorithms
for ECG signal Enhancement," International Journal of
Advanced Research in Compuer and Communication
Engineering, vol. 1, no. 2, pp. 86-90, April 2012.
[6] Rafi Ahamed Shaik, D V Rama Koti Reddy Md. Zia
Ur Rahman, "Noise Cancellation in ECG Signals
using Computationally Simplified Adaptive Filtering
Techniques: Application to," An International Journal
(SPIJ) , vol. 3, no. 5, pp. 120-131.
[7] U V Ratna Kumari G Sundeep, "Reduction of Power
Line Interference by Using Adaptive Filtering
Techniques”, International Journal of Innovative
Technology and Exploring Engineering, vol. 1, no. 5,
pp. 83-86, october 2012.
[8] Johann hauer olga shultseva, "Implementation of
adaptive Filters for ECG data Processing," in IEEE
Region 8 Sibicorn, 2008, pp. 206-209.
[9] Yunfeng WU,Rangaraj M.Rangaygen,Ye wu, and Sin-
Chun Ng, “Filtering of noise in Electrocardiographic
signal using an unbiased and normalized adaptive
artifact cancellation system” proceedings of HFSI &
IC FBI 2007, Hang Zhou,China October 12-14,2007.
[10] Maryam Afzali, Bijan Vosoughi Vahdat Mina Amiri,
“Comparison of Different Electrocardiogram Signal”
in precedings of the 19th Iranian Conference on
Biomedical Engineering (ICBME), Tehran, 2012, pp.
159-162.
[11] Rex Y. P. Cheung and Mark P. C. Mok Yuzhong Jiao,
“A Modified Log-LMS Adaptive Filter with Low Signal
Distortion” in 34th Annual International Conference of
the IEEE EMBS, san diego USA, 2012, pp.
21.67
19.04 19.76
21.86 22.17
0.148 0.170 0.114
1.136
1.506
3000
1000
750 800
500
0
500
1000
1500
2000
2500
3000
3500
0
5
10
15
20
25
LMS NLMS SDLMS LLMS MNLMS
Connvergence
SNR
Adaptive Filters
SNR
Time Complexity
Convergence

A novel adaptive algorithm for removal of power line interference from ecg signal

More Related Content

What's hot (18)

Similar to A novel adaptive algorithm for removal of power line interference from ecg signal (20)

More from International Journal of Computer and Communication System Engineering (20)

Recently uploaded (20)

A novel adaptive algorithm for removal of power line interference from ecg signal