International Journal of Current Engineering and Technology

ISSN 2277 - 4106

© 2013 INPRESSCO. All Rights Reserved.
Available at http://inpressco.com/category/ijcet

Research Article
Analysis of Adaptive Filter Algorithms using MATLAB
P Yadava, KP Gowdb, P.S. Singhelb, A Kharec and SK Paranjpeb*
a
All Saint College of Technology Bhopal, India -462031
b
AISECT University, Bhopal-Chiklod Road, Raisen, Bhopal, India
c
Dept of Electronics and communication, UIT, RGPV, Bhopal, India - 462031

Accepted 10 August 2013, Available online 25 August 2013, Vol.3, No.3 (August 2013)

Abstract

In recent years, adaptive filtering has become one of the effective and popular approaches for the processing and
analysis of the signals with noise especially of the biomedical signals. Adaptive filters permit to detect time varying
potentials and to track the dynamic variations of the signals. Besides, they modify their behavior according to the input
signal. Therefore, they can detect shape variations in the ensemble and thus they can obtain a better signal estimation.
The aim of this paper is to study, analyze various adaptive filter algorithms and apply Mat lab to investigate their
performance behaviors with two step sizes of 0.02 and 0.04. Further to remove motion artifacts from Electrocardiogram
signal as an application of this concepts. At the end of this paper, a performance study has been done between these
algorithms based on various step sizes. It has been found that there will be always tradeoff between step sizes and Mean
square error. The Electrocardiogram signals used in this paper are from the MIT-BIH database. Elimination of noises
from Electrocardiogram signal example is a classical problem.

Keywords: Adaptive filter, Least Mean Square (LMS), Normalized LMS (NLMS), Block LMS (BLMS), Sign LMS (SLMS),
Signed Regressor LMS (SRLMS), Motion artifact.

1. Introduction noise signal n(n) from corrupted signal d(n). However

entire noise source is difficult to obtain, estimated noise
1
A system is said to be adaptive when it tries to adjust its signal n’ (n) is used. The estimate noise signal n’ (n) is
parameters with the aid of meeting some well- defined calculated through some filters and measurable noise
goal or target that depends upon the state of the system source X(n) which is linearly related with noise signal
and its surroundings. So the system adjusts itself so as to n(n). After that, using estimated signal n’ (n) and obtained
respond to some phenomenon that is taking place in its signal d (n), estimated desired signal s’ (n) can be
surroundings. An event related signal could be considered obtained. If estimated noise signal n’ (n) is more close to
as a process, which can be decomposed into an invariant real noise signal n(n), then more desired signal is obtained.
deterministic signal time locked to a stimulus and an
additive noise uncorrelated with the signal. The most
common signal processing of this type of bioelectric signal
separates the deterministic signal from the noise. Several
techniques can be considered of which we are considering
the adaptive signal processing technique. Adaptive filters
are self-designing filters based on an algorithm which
allows the filter to learn the initial input statistics and to
track them if they are time varying. These filters estimate
the deterministic signal and remove the noise uncorrelated Fig.1 The principle of adaptive filter
with the deterministic signal. The principle of adaptive
filter is as shown in Figure 1. In the active noise cancellation theory, adaptive filter is
Obtained signal d (n) from sensor contains not only used. Adaptive filter is classified into two parts, adaptive
desired signal s (n) but also undesired noise signal n (n). algorithm and digital filter. Function of adaptive algorithm
Therefore measured signal from sensor is distorted by is making proper filter coefficient. General digital filters
noise n (n). At that time, if undesired Noise signal n(n) is use fixed coefficients, but adaptive filter change filter
known, desired signal s(n) can be obtained by subtracting coefficients in consideration of input signal, environment,
and output signal characteristics. Using this continuously
changed filter coefficient, estimated noise signal n’ (n) is
*Corresponding author: SK Paranjpe

1130
P Yadav et al International Journal of Current Engineering and Technology, Vol.3, No.3 (August 2013)

made by filtering X (n). The different types of adaptive coefficients are held constant over each block of L
filter algorithms can be explained as follows. samples, and the filter output y(n) and the error e(n) for
each value of n within the block are calculated using the
(a) LMS Algorithm filter coefficients for that block. Then at the end of each
block, the coefficients are updated using an average for the
The LMS algorithm is a method to estimate gradient L gradients estimates over the block.
vector with instantaneous value. It changes the filter tap
weights so that e (n) is minimized in the mean- square (e) Normalized LMS Algorithm (NLMS)
sense. The conventional LMS algorithm is a stochastic
implementation of the steepest descent algorithm. It In NLMS, the step size takes the form of,
simply replaces the cost function
ξ (n) = E [e2 (n)] (7)
by its instantaneous coarse estimate. The error estimation
e(n) is Where β is a normalized step size with 0< β<2. When x(n)
is large, the LMS experiences a problem with gradient
e (n) = d(n) – w(n) X(n) (2) noise amplification. With the normalization of the LMS
Coefficient updating equation is step size by ||x(n)||2 in the NLMS, noise amplification
problem is diminished.
w (n+1) = w(n) + μ x(n) e(n), (3)
2. Methodology
Where μ is an appropriate step size to be chosen as 0 < μ <
0.2, for the convergence of the algorithm. The larger step Electrocardiogram is a method of monitoring and
sizes make the coefficients to fluctuate wildly and recording the electric currents generated during the
eventually become unstable. The most important members alternating contractions of the atria and ventricles of the
of simplified LMS algorithms are: heart. The device used to monitor and record these signals
is an electrocardiogram more commonly referred to as an
(b) SRLMS Algorithm Electrocardiogram. When using an Electrocardiogram,
electrodes are applied to the skin in places where the
The signed regressor algorithm is obtained from the heart’s signals can be measured easily. Cables connect the
conventional LMS recursion by replacing the tap- input electrodes to the Electrocardiogram where the electrical
vector x (n) with the vector sgn{x(n)} .Consider a signed signal is turned into a waveform on a computer or a paper
regressor LMS based adaptive filter that processes an plot. The results produced from this machine allow
input signal x(n) and generates the output y(n) as per the physicians to observe the performance and condition of
following: the heart as well as diagnose any problems they may find
in the signal. When the doctors are examining the patient
y (n) = wt (n)x(n) (4) on-line and want to review the Electrocardiogram signal
waveform in real-time, there is a good chance that the
where, w(n) = [ w0(n), w1(n), ... , wL-1(n) ]t is a L-th signal has been contaminated by baseline wander (BW),
order adaptive filter. The adaptive filter coefficients are power line interference (PLI), muscle artifacts (MA) and
updated by the Signed-regressor LMS algorithm as, electrode motion artifacts (EM) etc., mainly caused by
patient breathing, movement, power line noise, bad
w (n+1) = w(n) + μ sgn{x(n)}e(n) (5) electrodes and improper electrode site preparation. All
these noises mask the tiny features of the signal and leads
Because of the replacement of x(n) by its sign, to false diagnosis. To allow doctors to view the best signal
implementation of this recursion may be cheaper than the that can be obtained, we need to develop an adaptive filter
conventional LMS recursion, especially in high speed to remove the artifacts in order to better obtain and
applications such as biotelemetry these types of recursions interpret the respiratory signal data.
may be necessary. In this proposed methodology simulation was carried
out using powerful MATLAB tool to investigate the
(c) SLMS Algorithm performance behaviors of various adaptive filter
algorithms in non-stationary environment with two step
This algorithm is obtained from conventional LMS sizes of 0.02 and 0.004. The principle means of
recursion by replacing e(n) by its sign. This leads to the comparison is the error cancellation capability of the
following recursion: algorithms which depends on the parameters such as step
w(n+1) = w(n) + μ x(n) sgn{e(n)} (6) size, filter length and number of iterations. A synthetically
generated motion artifact is added with Electrocardiogram
(d) Block LMS Algorithm (BLMS) signals. It is then removed using adaptive filter algorithms
such as LMS, Sign LMS, Signed Regressor, BLMS and
To reduce the computational requirements of LMS NLMS. All Simulations presented are averages over 1000
algorithm, block LMS is introduced. Here the filter independent runs.

1131
P Yadav et al International Journal of Current Engineering and Technology, Vol.3, No.3 (August 2013)

Motion Artifact Electrocardiogram. Basic concepts of adaptive filter

algorithms and mathematical support for all the algorithms
A number of signal-distorting events are noteworthy. are taken from (Monson Hayes H,2002).
Electrocardiograms operate best when the subject is in The INLMS algorithm extends the gradient-adaptive
resting position. Motion artifacts are unwanted signal and learning rate approach to the case where the signals are
are generally caused by relative motion between electrode nonstationary. It is shown that the INLMS algorithm can
and skin surface. When the skin stretches, the voltage work even for highly nonstationary interference signals,
changes which is recorded by the electrodes and summed where previous gradient- adaptive learning rate algorithms
with the Electrocardiogram signal. Pushing and pulling on fail. The use of two simple and robust variable step-size
the lead wires and patient movement are the most approaches in the adaptation process of the Normalized
recurring causes change in charge boundary at the Least Mean Square algorithm in the adaptive channel
electrode-electrode paste interface or electrode-skin equalization is investigated in (S. A. Jimaa et al,2007).
interface can also cause motion artifact In the Convergence Evaluation of Variable Step-Size
NLMS Algorithm in Adaptive Channel Equalization, the
3. Results and discussions input power and error signals are used to design the step
size parameter at each iteration. Simulation results
The extraction of high-resolution ECG signals from demonstrate that in the scenario of channel equalization,
recordings infected with back ground noise is an important the proposed algorithm accomplishes faster start-up and
issue to investigate. The goal for ECG signal enhancement gives better precision than the conventional algorithms. A
is to separate the valid signal components from the novel power-line interference (PLI) detection and
undesired artifacts, so as to present an ECG that facilitates suppression algorithm is presented in (Hideki et al,2008)to
easy and accurate interpretation. Many approaches have preprocess the electrocardiogram signals. A distinct
been reported in the literature to address ECG feature of this proposed algorithm is its ability to detect
enhancement (N. V. Thakor et al,1991). In recent years, the presence of PLI in the Electrocardiogram signal before
adaptive filtering has become one of the effective and applying the PLI suppression algorithm. An efficient
popular approaches for the processing and analysis of the recursive least-squares (RLS) adaptive notch filter is also
ECG and other biomedical signals. Adaptive filters permit developed to serve the purpose of PLI suppression. In
to detect time varying potentials and to track the dynamic (Yue-Der Lin et al,2008)two types of adaptive filters are
variations of the signals. Besides, they modify their considered to reduce the Electrocardiogram signal noises
behavior according to the input signal. Therefore, they can like PLI and Base Line Interference. Various methods of
detect shape variations in the ensemble and thus they can removing noises from Electrocardiogram signal and its
obtain a better signal estimation, quite a few papers have implementation using the Lab view tool was referred in
been presented in the area of biomedical signal processing (Yue-Der Lin et al,2008). Results in (Tutorial in Labview)
where an adaptive solution based on the LMS algorithm is indicate that respiratory signals alone are sufficient and
suggested (Allan Kardec Barros et al,1997). The perform even better than the combined respiratory and
fundamental principles of adaptive filtering for noise Electrocardiogram signals.
cancelation were described by Widrow et al. Thakor and
Zhu proposed an adaptive recurrent filter to acquire the Removal of Motion Artifacts
impulse response of normal QRS complexes, and then
applied it for arrhythmia detection in ambulatory ECG Respiratory signal is represented by second-order
recordings. The reference inputs to the LMS algorithm are autoregressive process that is generated according to the
deterministic functions and are defined by a periodically difference equation
extended, truncated set of orthonormal basis functions. In
these papers, the LMS algorithm operates on an x(n)=1.2728x(n-1) – 0.81x(n-2) + v(n) (8)
instantaneous basis such that the estimate. In a recent
study, however, a steady state convergence analysis for the Where v (n) is randomly generated noise
LMS algorithm with deterministic reference inputs
showed that the steady-state weight vector is biased, and
thus, the adaptive estimate does not approach the Wiener
solution. To handle this drawback another strategy was
considered for estimating the coefficients of the linear
expansion, namely, the block LMS (BLMS) algorithm , in
which the coefficient vector is updated only once every
occurrence based on a block gradient estimation. A major
advantage of the block, or the transform domain LMS
algorithm is that the input signals are approximately
uncorrelated. Mean square error behavior, convergence
and steady state analysis of different adaptive algorithms
are analyzed by S.C.Chan et al. The results of Desmond B Fig.2. LMS adoptive filter co-efficient plot of trajectories
show the performance analysis of adaptive filtering for for μ=0.02

1132
P Yadav et al International Journal of Current Engineering and Technology, Vol.3, No.3 (August 2013)

Figures 2, 3, 4, 5 and 6, 7, 8, 9 show the convergence of

filter coefficients and Mean squared error using LMS and
NLMS algorithms. An FIR filter order of 32 and adaptive
step size parameter (μ) of 0.02 and 0.004 are used for
LMS and modified step sizes (β) of 0.01 and 0.05 for
NLMS. It is inferred that the MSE performance is better
for NLMS when compared to LMS. The merits of LMS
algorithm is less consumption of memory and amount of
calculations.

Fig.6. NLMS adoptive filter co-efficient plot of

trajectories for μ=0.02

Fig.3. LMS adoptive filter squared error plot of

trajectories for μ=0.02
Fig.7. NLMS adoptive filter squared error plot of
trajectories for μ=0.02

Fig.4. LMS adoptive filter co-efficient plot of trajectories

for μ=0.004
Fig.8. NLMS adoptive filter co-efficient plot of
trajectories for μ=0.004

Fig.5. LMS adoptive filter squared error plot of Fig.9. NLMS adoptive filter squared error plot of
trajectories for μ=0.004 trajectories for μ=0.004

1133
P Yadav et al International Journal of Current Engineering and Technology, Vol.3, No.3 (August 2013)

4. Comparative assessment of results LMS. The merits of LMS algorithm is less consumption of
memory and amount of calculation. It has been found that
Table 1 provides the comparison of mean squared error there will be always tradeoff between step sizes and Mean
(MSE) and Convergence rate (C in terms of number of square error. It is also observed that the performance
iterations that the filter coefficients converge) of different depends on the number of samples taken for consideration.
algorithms. It is observed from Figure 2 and Figure 3, the
convergence speed for μ =0.02 is faster than μ=0.004. But Table 2: Comparison of MSE in removing motion
MSE performance is comparatively better for μ=0.004 artifacts
than μ=0.02. Convergence rate of LMS algorithm is better
when μ=0.02 and low MSE value when μ=0.004. It is also Algorithm Motion Artifacts
inferred that the MSE performance of Sign Regressor
LMS (SRLMS) at the step size of 0.02 is better when
compared to other algorithms. But there is always tradeoff μ=0.02 μ=0.004
between convergence rate and mean squared error. Hence MSE MSE
choosing an algorithm depends on the parameter on which LMS 1 .6e-007 2.66e-005
system has more concern. BLMS 3.2e-004 0.016
Table 1: Comparison of MSE and Convergence Rate SR LMS 5.4e-007 2.153e-007
SIGN LMS 2.0e-007 1.213e-005
Algorithm µ=0.02 µ=0.004 SIGN-SIGN LMS 3.5e-006 5.559e-007
NLMS β = 0.05 β = 0.01
MSE C MSE C 2. 1e-007 1 .057e-008
LMS 2.3873e 10 5.4907e 25
-004 0 -005 0 Conclusion
SRLMS 8.5993e 10 5.3036e 55 This study has revealed useful properties of various
-006 0 -004 0 adaptive filter algorithms. The objective is to optimize
SLMS 1.3406e 10 4.9436e 55 different adaptive filter algorithms so that we can reduce
-004 0 -005 0 the MSE so as to improve the quality of eliminating
interference. It is inferred that the MSE performance is
BLMS 4.9514e 20 8.7072e 50
better for NLMS when compared to LMS. The merits of
-004 0 -004 0 LMS algorithm is less consumption of memory and
NLMS β=0.05, 10 β=0.01, 70 amount of calculation. It has been found that there will be
6.8306e 0 0.0012 0 always tradeoff between step sizes and Mean square error.
-004 It is also observed that the performance depends on the
number of samples taken for consideration. Choosing an
algorithm depends on the parameter on which the system
Table 2 shows the comparison of resulting mean square has much concern. The future work includes the
error while eliminating Motion Artifacts from respiratory optimization of algorithms for all kinds of noises and to
signals using various adaptive filter algorithms with use the optimized one in the implementation of DSP
different step sizes. The observed MSE for LMS as shown Microcontroller that estimates the respiratory signal.
in Figure 5 (a) is very low for μ =0.02 compared with μ
=0.004. The performance of BLMS depends on block References
length L and NLMS depends on the normalized step size
β. Observing all cases, we can infer that choosing μ =0.02 B. Widrow, J. Glover, J. M. McCool, J. Kaunitz, C. S. Williams,
for the removal of power line interference is better when R. H.Hearn, J. R. Zeidler, E.Dong, and R. Goodlin (Dec
compared to μ =0.004. The step size μ =0.004 can be used 1975),Adaptive noise cancelling: Principles and applications ,
unless the convergence speed is a matter of great concern. Proc. IEEE, vol.63, pp.1692-1716.
It is found that the value of MSE also depends on the A. K. Barros and N. Ohnishi (Sept 1997),MSE behavior of
number of samples taken for analysis. biomedical event-related filters, IEEE Trans. Biomed. Eng.,
From the simulation results, the proposed adaptive vol. 44, pp. 848-855.
filter can support the task of eliminating motion artifacts O. Sayadi and M. B. Shamsollahi (Jan 2008),Model-based
fiducial points extraction for baseline wander
with fast numerical convergence. The mean square value electrocardiograms, IEEE Trans. Biomed. Eng., vol. 55, pp.
obtained in this work is found to be very low by varying 347-35 1.
the step sizes and increasing the number of iterations. An Y. Der Lin and Y. Hen Hu (Jan 2008),Power-line interference
FIR filter order of 32 and adaptive step size parameter (μ) detection and suppression in ECG signal processing, IEEE
of 0.02 and 0.004 are used for LMS and modified step Trans. Biomed. Eng., vol. 55, pp. 354-357.
sizes (β) of 0.01 and 0.05 for NLMS. It is inferred that the N. V. Thakor and Y.-S. Zhu (1991),Applications of adaptive
MSE performance is better for NLMS when compared to filtering to ECG analysis: noise cancellation and arrhythmia

1134
P Yadav et al International Journal of Current Engineering and Technology, Vol.3, No.3 (August 2013)

detection, IEEE Transactions on Biomedical Engineering, vol. Author’s Information

38,no. 8, pp. 785-794.
Ziarani. A. K, Konrad. A (2002),A nonlinear adaptive method of Prof. P Yadav is a post graduate from IIT- Kanpur
elimination of power line interference in ECG signals, IEEE (India) and a research scholar in Electronics Engg.
Transactions on Biomedical Engineering, Vol49, No.6, He has served in Indian Air Force as commissioned
officer for 25 yrs, and nine yrs, as a technocrat
pp.540-547.
academician in reputed Engg. Institutes. He is an
S. Olmos , L. Sornmo and P. Laguna (May 2002),Block adaptive entrepreneur who has many project patents under
filter with deterministic reference inputs for event-related his name. He is a fellow of IETE.
signals:BLMS and BRLS, IEEE Trans. Signal Processing, vol.
50, pp. 1102-1112.
Allan Kardec Barros and Noboru Ohnishi (Sept 1991), MSE
Behavior of Biomedical Event-Related Filters IEEE K Prandham Gowd obtained his B.Tech in
Transactions on Biomedical Engineering, 44(9). Electronics and Communication Engineering with
distinction from S.V. University, Tirupati (India)
Ahmed I. Sulyman, Azzedine Zerguine (2003), Convergence and
and ME (Microwaves and Radar) from IIT Roorkee.
Steady-State Analysis of a Variable Step-Size Normalized In 1994 he has conducted RCS Reduction
LMS Algorithm, IEEE. experiments on coated (by pasting of absorber
S.C.Chan, Z.G.Zhang, Y.Zhou, and Y.Hu (2008), A New Noise- sheets) and uncoated scaled models of aircraft which
Constrained Normalized Least Mean Squares Adaptive is first time in India at IIT Roorkee. He has 32
Filtering Algorithm, IEEE. research publications and 05 Technical reports to
Desmond B. Keenan, Paul Grossman (2006), Adaptive Filtering his credit most of them on RCS/RCS Reduction/Stealth Technology.
of Heart Rate Signals for an Improved Measure of Cardiac He has one copyright to his credit on Dynamic RCS Range Validation
Procedure from Govt of India. He is a Life Member of All India
Autonomic Control. nternational Journal of Signal
Management Association (AIMA), AeSI and Fellow of IETE. He had
Processing. authored a book on Stealth Aircraft Technology.
Monson Hayes H. (2002),Statistical Digital Signal Processing
and Modelling – John Wiley & Sons.
S. A. Jimaa, A.Simiri, R. M. Shubair, and T. Shimamura (Dec Pramil Singh Received the B.E. Degree in
2007), Collings, Interference-Normalized Least Mean Square Electronics & communication engineering from
Algorithm, IEEE Signal Processing Letters, 14(12). Rajiv Gandhi Proudyogiki Vishwavidyalaya,
Convergence Evaluation of Variable Step-Size NLMS Algorithm Bhopal, India in 2006 and M.Tech. Degree in
Microelectronics & VLSI Design engineering from
in Adaptive Channel Equalization IEEE 2009.
S.G.S.I.T.S. Indore, India in 2010.
Hideki Takekawa,Tetsuya Shimamura and Shihab Jimaa (2008),
An Efficient and Effective Variable Step Size NLMS
Algorithm IEEE.
Yue-Der Lin and Yu Hen Hu (Jan 2008), Power-Line Dr Anubhuti Khare received her BE in
Interference Detection and Suppression in ELEC Electronics and Communication from
TROCARDIOGRAM Signal Processing IEEE Transactions Government Engineering College, Bhopal in
on Biomedical Engineering, 55(1). 1994. She obtained her M.Tech and Ph D in
Electronics and Communication from MANIT,
Sachin singh and Dr K. L. Yadav (2010) , Performance
Bhopal. Presently she is working as Associate
Evaluation Of Different Adaptive Filters For Professor in E & C Department, UIT, RGPV,
ELECTROCARDIOGRAM Signal Processing ,International Bhopal. She has more than 50 publications to her
Journal on Computer Science and Engineering 02,(05). credit.
0Tutorial on Labview for ELECTROCARDIOGRAM Signal
Processing developed by NI Developer Zone, National
Instruments, April 2010, [online] Available at: Dr SK Paranjpe obtained his BE(Hons) in
http://zone.ni.com/dzhp/app/main Telecommunication Engg from Govt Engineering
College, Jabalpur and ME in Advanced
Walter Karlen, Claudio Mattiussi, and Dario Floreano (April
Electronics from IISc Bangalore in 1965 and
2009), Sleep and Wake Classification With 1967 respectively. He did his PhD in Computer
ELECTROCARDIOGRAM and Respiratory Effort Signals”, Hardware/Parallel Processing from University of
IEEE Transactions on Biomedical Circuits and Systems, 3(2). Roorkee (Now IIT-R) in 1985. Formerly he was
Director, Directorate of Technical Education,
Govt of Madhya Pradesh.

1135