[EEE/OSA/[APR International Conference on [nfonnatics, Electronics & Vision

Least-Mean-Square Algorithm Based Adaptive Filters

for Removing Power Line Interference from ECG
Signal

Md. Maniruzzaman, Kazi Md. Shimul Billah, Uzzal Biswas, and Bablu Gain
Electronics and Communication Engineering Discipline
Khulna University, Khulna 9208, Bangladesh
-

Abstract-The 50 Hz power line is one of the main sources of Adjoint LMS, Filtered XLMS, NLMS and FFT BLMS. We
interference in ECG signal measurement, and it distorts the have used the Signal Processing Toolbox of the mentioned
original ECG signal while recording. Recently, adaptive filtering ®
algorithms built in MATLAB . We also have evaluated their
has become one of the effective and popular methods for the performance. [t reveals that among all the adaptive filters, the
processing and analysis of the ECG signal. In this study, we have adaptive NLMS filter removes the 50 Hz power line
used adaptive filters to remove the power line interference from
interference more effectively.
the ECG signal. We have used different adaptive filter algorithms,
such as, Least-Mean-Square (LMS), Block LMS (BLMS), Delay
LMS (DLMS), Adjoint LMS, Filtered-X (XLMS), Normalized [I. MATERIALS AN D METHODS
LMS (NLMS) and Fast Fourier Transform BLMS (FFT BLMS). ®
The original ECG signal is generated by using MATLAB .
We have used the Signal Processing Toolbox of the mentioned The 50 Hz power line interference is also generated by using
® ®
algorithms built in MATLAB . It reveals that among all the
MATLAB . This 50 Hz noise signal is then added with the
adaptive filters, the adaptive NLMS filter removes the 50 Hz
original ECG signal. In order to remove the 50 Hz power line
power line interference more effectively.
interference and to reconstruct the ECG signal from the mixed
signal we use different adaptive filter algorithms. However, the
Keywords-power line interference; EeG signal; adaptive NLMS
basic block diagram for understanding the overall adaptive
filter; SNR
filtering process is depicted in Fig. 1.

I. IN TRODUCTION
den) X(n) + N(n)
=
e(n) = X'(n)
The ECG signal is extremely important for the diagnosis of
the cardiac patients. However, when the ECG signal is
recorded, it may be corrupted by various kinds of noises, such
as, power line interference, base line wandering, electrode
'(n)
contact noise, motion artifacts, muscle contraction,
instrumentation noise generated by electronic devices and
electrosurgical noise, etc. During the diagnosis of arrhythmia
or myocardial infraction, the 50 Hz power line noise can affect
the ECG signal. The frequency range of ECG signal is
generally 0.05 Hz to 100 Hz, and, that of the power line
interference is 50 Hz which lies in the ECG signal band. So, it
has become very crucial to remove the power line interference Figure I, Principle of adaptive filter.
from the ECG signal. Different types of digital filters (FIR and
IIR) have been used to solve the problem [I ]-[5]. However, it is As seen in Fig. 1, the mixed signal den) is obtained by
difficult to apply these filters with fixed coefficients to reduce adding the original ECG signal X(n) and the 50 Hz noise signal
the power line interference, because the ECG signal is known N(n). In reality, it is difficult to estimate the noise signal
as a non-stationary signal. Recently, adaptive filtering has because of their harmonics. [f we knew the value of noise
become one of the effective and popular methods for the signal, we could get back the original signal by simply
processing and analysis of the ECG signal [6]-[8]. It is well­ subtracting the noise signal from the mixed signal. To
known that adaptive filters with LMS algorithm show good overcome this problem, adaptive algorithm is used and an
performance for processing and analysis of the most of the estimated noise signal N'(n) is assumed. This estimated noise
biomedical signals which are non-stationary [6]. And in this signal is calculated through some filters and measureable noise
study, we have used adaptive filters to remove the power line source Sen) which is linearly correlated with the noise signal
interference from the ECG signal. We have used different N(n). By subtracting this estimated noise signal N'(n) from the
adaptive filter algorithms, such as, LMS, BLMS, DLMS, mixed signal den), we obtain the estimated desired signal X'(n)
which is also called the reconstructed signal.

978-1-4673-1154-0112/$31.00 ©2012 IEEE ICIEV 2012

 IEEE/OSA/IAPR International Conference on Infonnatics, Electronics & Vision

Then the reconstructed signal's power, signal-to-noise ratio

(c)
(SNR), percentage root mean square difference (%PRD) and
error standard deviation (ESD) are calculated to measure the � Original ECG Signal

performance of different adaptive filters.

J�
Moreover, a pragmatic step size ([.1) is used for algorithm � 1
- .0
updating and determining both how quickly and how closely i 1 mrr�mmTImrrm mrrmrrmmTImrrmillrmrrmrnmmrrmrrmmrrrnmmTIm
r
the adaptive filter adapts to the filter solution.
�
� 0
I ·'0������ �������
� Mixed Signal
III. RESUL IS AND DISCUSSION
J:
i : �;�O�:;:�O�
The 4-beat original ECG signal is generated by using
®
MATLAB whose sampling frequency is 500 Hz for each beat � 2
- 0 200 400 600 800 1000 1200 1400 1600 1800 2000

and amplitude is 1 mY. The 50 Hz power line interference is

also generated with sampling frequency of 2000 Hz. The power
line interference is then added to the original ECG signal to get � - 0
2
200 400 600 800 1000 1200 1400 1600 1800 2000
Sample index
the mixed signal. Finally, the power line interference is
removed using different adaptive filters based on different
algorithms, such as, LMS, BLMS, DLMS, Adjoint LMS,
Filtered XLMS, NLMS and FFT BLMS. The results are shown Figure 2. (c) Power line interference removing using adaptive D LMS filter for
in Figs. 2 (a) - 2 (g). /l = 0.008 and delay = 1.

(a) (d)

i: i:
f 1
- 0 200 400 600 800 1000 1200 1400 1600 1800 2000 f 1
- 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Noise Signal
> 1 mnrrmmnmmmmTImrrmmrrmnmmTImnmrrmrm� > 1
E E
� 0 � 0
�.l: · ' �llillllllilliillliillliW
0 200 400
WillllWllillllil
llili�wmliWllill
l llillilllllli
lliW�
il r

i: i:
f 2
�

i :�F':'�;" 3:j
2 - 0 200 400 600 800 1000 1200 1400 1600 1800 2000
- 0 200 400 600 800 1000 1200 1400 1600 1800 2000

� � 3:j
Reconstructed ECG Si9,"" '

�
� 1 __
�_�--;,-�_,
0 ·. :'�:"

� 1
- 0 200 400 600 800 1000 1200 1400 1600 1800 2000 t 1
- 0 200 400 600 800 1000
Sample index
1200 1400 1600 1800 2000
Sample index

Figure 2. (a) Power line interference removing using adaptive LMS filter for Figure 2. (d) Power line interference removing using adaptive adjoint LMS
filter tap = 32 and /l = 0.008. filter for filter tap = 32 and /l = 0.008.

(b) (e)
� Original ECG Signal
Original ECG Signal

I:
'
�
0

t -
1 00 �� 1
- 0
�
200 400 600 800 1000 1200 1400 1600 1800 2000
i 1 mn�����rrm�mm� ��mrrmr
mmmmn=TImrrmm
r 1 mTIrrmmrrmmmmmmTImrrm���mrnTImTImrrmmrrmmmmmrn�
J 0 ��

� �
0

I r ·' �
O =W� ���
� Mixed Signal Mixed Signal

I:
t -
2
00
I: f 2
- •0

.----- � \ � �';' �
Reconstructed ECG Signal

j�Hr=::fO�:;:O�
t 0 200 400 600 800 1000
Sample index
1200 1400 1600 1800 2000
� :�
� l 1
- 0 200 400 600 800 1000
Sample index
1200 1400 1600 1800 2000

Figure 2. (b) Power line interference removing using adaptive BLMS filter for Figure 2. (e) Power line interference removing using adaptive Filtered XLMS
filter tap = 32 and block length = 2. filter for filter tap = 32 and /l = 0.008.

738 ICIEV 2012

 IEEE/OSA/IAPR International Conference on Infonnatics, Electronics & Vision

(1) (g)

i�
Original ECG Signal

� -1
0 200 400 600 800 1000 1200 1400 1600 1800 2000
it �-
1
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Noise Sir;Jnal

� 1 ��������� ���������
� 0
t 1
- 0

i 2

��
0

� it :
i � �F:2:0 =:E
-2
0 200 400 600 800 1000 1200 1400 1600 1800 2000
-2

i :�:;I;�,:O �
00

f
t
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample index
-
1
� 8Do �OO
1
Sample index

Figure 2. (1) Power line interference removing using adaptive N LMS filter Figure 2. (g) Power line interference removing using adaptive FFT B LMS
for J.l = I and offset = 1 16 .. filter for J.l = 0.008 and block length = 25.

If the amplitude of the reconstructed signal increases, then

there will be high distortion and vice versa. When we take the TAB LE!. VALUES OF PERFORMANCE PARAMETERS OF DIFFERENT
ADAPTIVE FILTERS.
value of fl equal to 0.008, then we see that some noises also
appear on the R and S peak compared with the value of fl equal Adaptive Reconstructed signal's
to 0.005. We can also see that when fl is equal to 0.005, then Filters
Power (dB ) SN R % PRD ESD
the reconstructed signal amplitude is less than the original
signal as well as all other measuring values, such as, the SNR LMS - 16.246 - 10.22 1 1.79 1 0.274
and %PRO decreases, as seen in Table I. So we can say that the
SNR for step size fl of 0.008 is better but exhibits some B LMS - 16.242 - 10.222 1.803 0.274

distortion and high %PRO. On the other hand, when fl is 0.005 D LMS - 16.240 - 10.220 1.8 10 0.274
then the SNR and %PRD is low but the reconstructed signal
contains low distortion. In fact, as seen in Fig. 2, there is some Adjoint LMS - 16.58 1 - 10.56 1 0.784 0.367

distortion in the reconstructed signal because of step size fl. X LMS - 16.500 - 10.480 1.026 0.369

Fig. 3 shows the comparison of the different reconstructed N LMS - 16.237 - 10.2 17 1.8 18 0.274
ECG signals as they are the output of different adaptive filters.
They indicate that the reconstructed ECG signal obtained from FFT B LMS - 16.267 - 10.247 1.728 0.278
the adaptive NLMS filter has less error than the other adaptive
filters and the 50 Hz power line interference is removed by this
filter, more effectively.

To realize the performance of different filters, let us

analyze the Table I, where we can see that the adaptive NLMS
filter has the highest reconstructed signal power of -16.237 dB
and SNR of -10.217. Also, the % PRO is of moderate level of
l.818. The adaptive NLMS filter consist the lowest ESD is
0.274 compared to the other adaptive filter algorithms.

IV. CON CLUSION

The results reveal that the LMS and NLMS are better than
the other adaptive filters. However, the obtained SNR of
reconstructed signal of LMS filter is lower than that of the
NLMS filter. So, the NLMS adaptive filter is more appreciable
for removing the power line interference from the ECG signal.
Sample N umber

Figure 3. Comparison of reconstructed signals of diflerent adaptive filters: (a)

Original ECG signal, (b) LMS filter, (c) B LMS filter, (d) DLMS filter, (e)
Adjoint LMS filter, (f) Filtered X LMS filter, (g) N LMS filter and (h) FFT
B LMS filter.

739 ICIEV 2012

 IEEE/OSA/IAPR International Conference on Infonnatics, Electronics & Vision

Reduction," World Scientific and Engineering Academy and Society

REFERENCES (WSEAS),Stevens Point, Wisconsin, USA, vol. 1, no. 157-474, pp. 47-
50,January 2004.
[ 1] E. T. Gar,C. Thomas and M. Friesen, "Comparison of Noise Sensitivity
ofQRS Detection Algorithms," IEEE Tran. Biomed. Eng., vol. 37, no.l, [6] D. V. R. K. Reddy, M. Z. U. Rahman, Y. Sangeetha, and N. S. Sudha,
pp. 85-98, January 1990. "B ase Line Wander and Power Line Interference Elimination from
Cardiac Signals Using a Novel LMS Algorithm B ased On Differential
[2 ] R Limacher' "Removal of Power Line Interference from the ECG
Inputs and Errors," Int. J. of Advanced Eng. & App!., pp. 187 - 19 1,
i
S gnal by an Adaptive Digital Filter," in Proc. of European Tel. Conf, January 20 1 1.
Garmisch-Part,pp. 300-309,M ay 21-23, 1996.
[7] G. M ihov, I. V. Dotsinsky, T. S. Georgieva, C. Levkov, R. Ivanov, I. K.
[3 ] 1. M ateo, C. Sanchez, A. Torres, R. Cervigon, and 1. 1. Rieta, "N eural
Daskalov, and 1. Christov, "Subtraction Procedure for Powerline
Network B ased Canceller for Power Line Interference in ECG Signals,"
Interference Removing from ECG Improvement for Non-multiple
Computers in Cardiology, vol. 35,pp. 1073-1076,April 2008.
Sampling," J. Med. Eng. Techno!., vo1.29, no.5, pp.238-243, August
[4 ] M . Kaur and B . Singh, "Power Line Interference Reduction in ECG 2005.
Using Combination of MA M ethod and IIR Notch Filter," Int. J. of
[8 ] L. Sommo, "Time-Varying Filtering for Removal of B aseline Wander in
Recent Trends in Eng., vol. 2,no. 6,pp. 125-129,November 2009.
Exercise ECGs," Computers in Cardiology, IEEE Computer Soc. Press,
[5] M. S. Chavan, R. Agarwala, M. D. Uplane, and M. S. Gaikwad, "Design pp. 145-148,September 23-26, 199 1.
of ECG Instrumentation and Implementation of Digital Filter for Noise

740 ICIEV 2012