MATLAB Simulation Comparison for Different Adaptive Noise Cancelation

Algorithms

Mostafa Guda, Safa Gasser and Mohamed S. El Mahallawy

Department of Electronics & Communications Engineering
Arab Academy for Science, Technology and Maritime Transport, Egypt
mostafa.guda@hotmail.com, safagasser@aast.edu, mahallawy@aast.edu

ABSTRACT biological and environmental resources. The

Electrocardiographic (ECG) signal can be
contaminated by diverse forms of noise: baseline
wander, 60 Hz power line interference, muscle
noise, and motion artifact. 60Hz power line
interference can be cancelled using two different
approaches; an adaptive filter or notch filters. The
adaptive filter essentially minimizes the mean-
squared error between a primary input, which is the
noisy ECG, and a reference input, which is either
noise that is correlated in some way with the noise
in the primary input or a signal that is correlated
only with ECG in the primary input. In this paper
we present a MATLAB simulation comparison
between different adaptive filter algorithms; Least
Mean Square (LMS), Normalised LMS (NLMS),
Variable Step size LMS (VSLMS), Recursive Least
Square (RLS) and Blind LMS. The comparison is
carried out in terms of both, MSE and the algorithm
convergence rate.
KEYWORDS
ECG;ANC;LMS;NLMS;VSLMS;RLS;MSE;Blind
LMS
1 Introduction
The Electrocardiographic (ECG) signal is the
physical interpretation of the electrical behavior
created by the heart muscles. This electrical
activity of the heart is captured over time by an
external electrode attached to the skin and
recorded by a device external to the body. The
ECG signal frequency ranges from 0.5 to 100
Hz [1].
The ECG signal is normally corrupted with
two major types of noise generated by
first type includes muscle contraction,
electromyographic (EMG) interference,
baseline drift, ECG amplitude modulation due
to respiration, and motion artefacts caused by
changes in the electrode-skin impedance with
electrode motion. The second type includes
power line interference, electrode contact noise,
instrumentation noise generated by electronic
devices used in signal processing,
electrosurgical noise, and radio-frequency
[2],[3]. One common type of noise that occurs
during ECG collection is power line
interference of 60Hz.
ECG noise cancellation has long puzzled the
research community. There has been a
tremendous amount of research on how to
cancel noise in an ECG signal.
In [4] the authors used an Adaptive Noise
Canceller (ANC) using the Least Mean Square
(LMS) algorithm. The LMS algorithm was
devised by Widrow and Hoff in 1959. The
advantage of using the algorithm in [4] is its
low computational complexity, while it lacks
the high convergence speed due to its fixed step
size.
The authors in [5], [6] introduced another ANC
based on the Normalized LMS (NLMS)
algorithm. This algorithm has a higher
convergence rate than the algorithm used in [4].
The NLMS algorithm is also based on a fixed
step size as the LMS.
In [7] the Variable Step size LMS was
simulated. This algorithm has better estimation
ability but it also has high computational
complexity compared to the algorithms used in
[4], [5] and [6].

ISBN:978-0-9891305-6-1 ©2014 SDIWC 68

 A new approach was experimented in [8]. The (3)
authors used the Recursive Least Square
algorithm (RLS) in the ANC to eliminate the
ECG noise. The main disadvantage of this 2.2 NLMS Algorithm:
algorithm is its very high computational
complexity, while its advantage is its fast In (3), we notice that the step size depends on
convergence rate. the eigen values which make the mathematical
In [9] the authors used an adaptive filter process more complex. To make the
without a reference signal to cancel the ECG calculations easier in the NLMS algorithm the
noise. Other notation for this technique is Blind step size depends on the autocorrelation
adaptive filter. between of input signal, where
In this paper we give a simulation comparison
of the previously mentioned algorithms to (4)
cancel the power line interference occurring at
60 Hz of an ECG signal. Thus, this paper is
divided as follows: in Section II we give an The weights are consequently updated such that
overview about the theory of adaptive [4-6]
algorithms used in the ANC. In Section III we
show the simulation results obtained for LMS, ( ) ( ) ( ) ( ) (5)
NLMS, VSLMS, RLS and Blind LMS
algorithms when used in cancelling noise
occurring in an ECG signal along with a 2.3 VSLMS Algorithm
comparison of the MSE obtained in each
algorithm. In Section IV we sum up our work As opposed to a fixed step size, the VSLMS
and conclude our paper. algorithm improves the step size calculation by
making it variable in each iteration, where
( ) ( ) ( ) ( ) (6)
2 Adaptive Algorithms
( ) ( ) ( ) (7)
In this section we briefly explain each
algorithm and how each one differs in terms of Where g(n) is a gradient function. Finally, the
its update equation computation. weights are updated such that
( ) ( ) ( ) (8)
2.1 LMS Algorithm 2.4 RLS Algorithm
All the previous algorithms process the signals
The LMS algorithm uses the following equation one sample at a time. The RLS algorithm
to update the filter coefficients [1-3], [10-12]. processes the signal in overlapping blocks of
. samples. This increases the computational
w(n+1)=w(n)+2µe(n)x(n) (1) complexity and makes it harder to be
implemented. The filter weights are updated as
Where x(n) is the input vector, w(n) is the follows [10]
vector containing the filter coefficients and e(n)
is the error function given by ( ) ( ) ( ) (9)
( ) ( ) (10)
( ) ( )
e(n)=d(n)-w(n)x(n) (2)
( ) ( ) ( ) ( ) (11)
( )
µ, the step size, depends on the eigen values of ( ) ( ) (12)
the input signal where ( )[ ( ) ( )]

ISBN:978-0-9891305-6-1 ©2014 SDIWC 69

 Where k(n) is a gain vector. We notice the level
of complexity of the mathematics involved.
Despite having good results, the complexity of
the algorithm makes it less usable and difficult
to implement.
amplitude when compared to that used in the
other algorithms simulation. The purpose of
this step is to show the convergence of the
algorithm and the noise cancelling step.

2.5 Blind LMS Algorithm

The Blind LMS algorithm is very useful for

applications where it is hard to obtain
information about the noise affecting the input
signal. The input signal is delayed to make it
possible to get an early version of the noise for
processing [1]. Then equations (1), (2) and (3)
are used to cancel the noise.

3 Simulation Results

For the analysis of the ANC, the ECG real time

Figure 2. MSE of ANC Using LMS Algorithm
data are obtained from [13]. The noise is known
to be sinusoidal with unknown amplitude and
Since all algorithms are iterative, the noise still
phase. A reference signal is considered, which
appears in the first 100 samples. It gradually
has a different amplitude as well as phase than
decreases until it is approximately cancelled
that of the noise. A filter of order 10 was
from the signal as seen in “Fig.1”. Another way
simulated using MATLAB, as the results
to analyze the ability of the algorithm to cancel
obtained from the simulation are represented
the noise is the MSE. As seen in “Fig.2”, the
below.
error in estimation at starts at its highest value
3.1 LMS Simulation Results
and then converges gradually to approximately
the zero level. The convergence rate is affected
by the noise level, the filter order and the
choice of the step size.

3.2 NLMS Simulation Results

In the NLMS algorithm, the step size is a

Figure 1. ANC Simulation Using LMS Algorithm
Worth mentioning that the reference signal used
in the LMS algorithm simulation is different in
function of the input signal rather than being a
function in its eigen values. The main
advantage is to reduce the computations time
and complexity of the algorithm.
We can see that “Fig.1” and “Fig.3” are almost
the same. The noise starts to vanish after the
first hundred samples. The NLMS MSE curve
starts at a value lower than that of the LMS
algorithm and also converges faster as seen in
“Fig.2” and “Fig.4”.

ISBN:978-0-9891305-6-1 ©2014 SDIWC 70

 Figure 3. ANC Simulation Using NLMS Algorithm

Figure 6. MSE of ANC Using VSLMS Algorithm

The VSLMS algorithm updates the step size in

a way that makes it variable between the
iterations. This helps the algorithm converge
faster and get a better estimation of the noise
signal.
Due to the step size being variable in this
algorithm, the noise is cancelled quicker than
the previous algorithms as seen in “Fig.5”. It is
also clear in “Fig.6” that both the convergence
rate and the MSE are better than those of
“Fig.2” and “Fig.4”.
Figure 4. MSE of ANC Using NLMS Algorithm
3.4 RLS Simulation Results
3.3 VSLMS Simulation Results

Figure 7. ANC Simulation Using RLS Algorithm

Figure 5. ANC Simulation Using VSLMS Algorithm

ISBN:978-0-9891305-6-1 ©2014 SDIWC 71

 Figure 8. MSE of ANC Using RLS Algorithm
In the RLS algorithm, we used a block of 10
samples for processing. In each iteration there
are 9 common samples being processed and one
new sample. This helps in achieving a near
correct estimation through processing the
sample more than once as seen in “Fig.7”. The
MSE becomes better and the algorithm
converges faster as noticed in “Fig.8”.
3.5 Blind LMS Simulation Results
In the Blind LMS algorithm, we made the
reference signal a version of the noise signal.
This made the convergence rate faster than that
of the LMS as seen in “Fig.9”. Consequently,
the error in estimating the noise signal is much
better as seen in “Fig.10”.
Figure 9. ANC Simulation Using Blind LMS Algorithm
ISBN:978-0-9891305-6-1 ©2014 SDIWC
Figure 10. MSE of ANC Using Blind LMS Algorithm
3.6 MSE Comparison
One of the methods to compare the
performance of two or more algorithms is the
MSE. The MSE curve shows how the error
function behaves with the increase in the
number of iterations. As seen in “Fig.11”, the
LMS algorithm has the highest starting value
for the MSE. It also has the slowest
convergence rate. The NLMS algorithm has
both, a better convergence rate and better MSE
values. The VSLMS algorithm shows good
improvement and better performance than the
previously mentioned algorithms. The RLS
algorithm is the quickest to converge and has
the best MSE values. The Blind LMS algorithm
shows better convergence rate than that of the
LMS algorithm.
Figure 11. MSE Comparison of ANC Algorithms
72
 4 Conclusion
In this paper we introduced the simulation
results and comparison of different algorithms
used in 60 Hz noise cancellation of an ECG
signal. The simulation was done using
MATLAB program.
The first simulated algorithm was the LMS
algorithm. This algorithm processes the given
input signal one sample at a time through using
a constant step size. It has the slowest
convergence rate and the highest MSE.
The second simulated algorithm was the NLMS
algorithm. The step size in this algorithm was a
function of the input signal. This modification
resulted in a faster convergence rate and a
better MSE when compared to the LMS
algorithm.
The third simulated algorithm was the VSLMS
algorithm. This algorithm introduced a variable
step size to the adaptive filter. This step size
would be a function of the step size in the
previous iteration. The step size value is also
limited by a certain range in order to ensure the
algorithm converges within the range of the
given data.
The fourth simulated algorithm was the RLS
algorithm. This algorithm is unique in its way
of processing the given input data. The data is
processed in blocks instead of one sample at a
time as was the case in the LMS algorithm. The
larger the overlap between the processed
blocks, the better the results will be. This
algorithm also has a constant step size. The
RLS algorithm has a better convergence rate
and also a better MSE compared to the LMS
algorithm.
The final simulated algorithm was the Blind
LMS. The difference in this algorithm is in the
reference signal introduced to the adaptive
filter. Instead of trying to get an estimate of the
noise signal, we define the reference signal to
be the noise signal. The input data is
consequently introduced to the filter with a
delay. This new adjustment results in a better
estimation and faster noise cancellation
compared to the ordinary LMS algorithm.
ISBN:978-0-9891305-6-1 ©2014 SDIWC
5 REFERENCES
[1] D. Bhoyar, and P. Singh, “ECG Noise Removal
using Adaptive Filtering”, International Journal of
Engineering Research and Applications (IJERA)
ISSN: 2248-9622, International Conference on
Industrial Automation and Computing (ICIAC- 12-
13th April 2014)
[2] S. A. Rehman, and R.R. Kumar, “Performance
Comparison of Adaptive Filter Algorithms for ECG
Signal Enhancement”, International Journal of
Advanced Research in Computer and
Communication Engineering Vol. 1, Issue 2, April
2012
[3] S. Anjana, and X. A. Mary, “Noise Cancellation
Using Adaptive Filters in FPGA”, Research Journal
of Recent Sciences, Vol. 3(ISC-2013), 4-8 (2014)
[4] B. Widrow, J. R. Glover JR., J. M. Mccool, J.
Kaunitz, C. S. Williams, R. H. Hean, J. R. Zeidler,
E. Dong JR., and R. C. Goodlin, “ Adaptive Noise
Cancelling: Principles and Applications”. Proc.
IEEE, Vol. 63, NO. 12, December 1975
[5] R. K. Thenua, and S. K. Agrawal, “Hardware
Implementation of Adaptive Algorithms for Noise
Cancellation”, International Journal of Information
and Electronics Engineering, Vol. 2, No. 2, March
2012
[6] G. Sundeep, U. V. Ratna Kumari, “Reduction of
Power Line Interference by Using Adaptive Filtering
Techniques in Electrocardiogram”, International
Journal of Innovative Technology and Exploring
Engineering (IJITEE) ISSN: 2278-3075, Volume-1,
Issue-5, October 2012
[7] M. Hutson, Acoustic Echo Cancellation Using
Digital Signal Processing, The School of
Information Technology and Electrical Engineering,
The University of Queensland, November 2003
[8] A. Singh, R. Mehra, “ Adaptive Filter for ECG
Noise Reduction Using RLS Algorithm”, Int.
[9] Journal of Engineering Research and Applications
ISSN : 2248-9622, Vol. 3, Issue 6, Nov-Dec 2013,
pp.1304-1308
[10] C. Chang, H. Ko, and K. Chang, “Cancellation of
High-Frequency Noise in ECG Signals Using
Adaptive Filter without External Reference”, 3rd
International Conference on Biomedical Engineering
and Informatics (BMEI 2010)
[11] D.C. Dhubkarya, A. Katara, and R.K. Thenua,
“Simulation of Adaptive Noise Canceller for an
ECG signal Analysis”, ACEEE Int. J. on Signal &
Image Processing, Vol. 03, No. 01, Jan 2012
[12] R. K. Thenua, and S. K. Agrawal, “Hardware
Implementation of Adaptive Algorithms for Noise
Cancellation”, International Journal of Information
and Electronics Engineering, Vol. 2, No. 2, March
2012
[13] G. Singh, K. Savita, S. Yadav, V. Purwar, “Design
of Adaptive Noise Canceller Using LMS
Algorithm”, International Journal of Advanced
Technology & Engineering Research (IJATER),
Volume 3, Issue 3, May 2013
[14] http://physionet.org/cgi-bin/atm/ATM/accessed on
May 30,2014
73