ENERGY COMPUTATION FOR BCI USING DCT AND MOVING AVERAGE WINDOW FOR NOISE SMOOTHENING

International Journal of Computer Science, Engineering and Applications (IJCSEA) Vol.2, No.1, February 2012
DOI : 10.5121/ijcsea.2012.2102 15
ENERGY COMPUTATION FOR BCI USING DCT AND
MOVING AVERAGE WINDOW FOR NOISE
SMOOTHENING
Ch.Aparna1
, Dr. J.V.R.Murthy2
,Dr. B.Raveendra Babu3
1
R.V.R&J.C College of Engineering, Guntur, A.P.
chaparala_aparna@yahoo.com
2
Dept. CSE, JNTU College of Engineering, Kakinada, A.P.
mjonnalagedda@gmail.com
3
DELTA Technology & Management Services Pvt. Ltd., Madhapur, Hyderabad, A.P.
rbhogapathi@yahoo.com
ABSTRACT
Brain computer interface (BCI) is a fast evolving field of research enabling computers and machines to be
directly controlled by the human neural system. This enables people with muscular disability to directly
control machines using their thought process. The brain signals are recorded using Electro-
encephalography (EEG) and patterns extracted so that the BCI system should be able to classify various
patterns of brain signal accurately to perform different tasks. The raw EEG signal contains different kinds
of interference waveforms (artifacts) and noise. Thus raw signals cannot be directly used for classification,
the EEG signals has to undergo preprocessing, to remove artifacts and to extract the right attributes for
classification. In this paper it is proposed to extract the energies in the EEG signal and classify the signal
using Naïve Bayes and Instance based learners. The proposed method performs well for the two class
problem in the multiple datasets used..
KEYWORDS
BCI, EEG, Butterworth filters, Moving average spencer window, Discrete cosine transform (DCT), IBL.
1. INTRODUCTION
A Brain computer interface (BCI) system works by recording the brain signals and applying
machine learning algorithms to classify the brain signals and performing a computer controlled
action. BCI finds application for physically impaired people as no peripheral muscular activity is
required. The most widely used method for signal acquisition is the electro-encephalography
(EEG). The reason for the popularity of EEG is due to the non-invasive way of acquiring the
brain signals and also it is safe, easy and cheap when compared to other methods. Most of the
existing application prototypes of BCI use EEG signals. Prototypes like “Thought Translation
Device” [1] which allows paralyzed patient to write sentences, spelling system [2], “virtual
keyboard” based on motor activity [3] are all EEG based BCI systems. Thus the role of EEG
processing is crucial in the development of BCI.

16
The user can control BCI by producing different brain activity patterns which is identified by the
system and translated into commands. BCI automatically estimates the class of data using
classification algorithm. Investigation and evaluation of different classification algorithms for
EEG-based BCI is available in literature [4, 5]. Translating brain activity into a computer
command is achieved using Regression [6] or classification [7] algorithms.
Bassani et al.,(2010) [8] proposed continuous wavelet transformation (CWT) and wavelet
coherence (WC) for feature extraction. Patterns were extracted in the time-frequency plane and
optimized. Using WC analysis, the CWT allows the representation of time-frequency patterns of
the signal's information content. The proposed feature extraction method was tested using the
Naive Bayes classifier with good results.
Rutkowski et al.,(2009)[9] extensively investigated to remove artefacts caused by muscular
interference due to eye movement during the EEG signal recording.The proposed method used
correlation of the oscillatory modes within multi channel signal in the Hilbert domain.The
proposed method was examined and found to decrease the unwanted noise introduced during
involuntary movement of the eye.
Lotte et al.,(2009) [10] use fuzzy logic to obtain the region of interest. The proposed model,
Fuzzy Region of Interest Activity (FuRIA) is based on inverse solutions and new concepts of
fuzzy region. The advantage of the proposed algorithm is its trainability and the ROI and fuzzy
frequency bank were able to distinguish different mental states. The proposed technique was
compared with two data sets from the BCI competitions. The evaluations assessed the impact of
hyper-parameters and the effects of fuzzification on the performance. The feature extracted using
FURIA was tested using Support Vector Machine (SVM) classifier.
In this paper it is proposed to extract maximum and average energy in the 5-30 Hz. To extract the
signals in the 5-30 Hz range and remove random noise and at the same time maintaining a sharp
step response we use two stages of filtering with the first stage being a butterworth filter and the
second stage consisting of an moving average 7 point spencer filter. The obtained features were
used as classification attributes for Naïve Bayes and Instance Based Learners(IBL) algorithm.
2. MATERIALS AND METHODS
Filters allow a specified frequency range to pass through while attenuating the specified unwanted
frequency range. Low-pass filter passes low-frequency signals reduces the amplitude of signals
with frequencies higher than the cut-off frequency. High-pass filter passes high frequencies and
attenuates frequencies lower than its cut-off frequency. A band-pass filter is a combination of a
low-pass and a high-pass filter; it passes a certain band of frequencies.
The Butterworth filter is a signal processing filter which is so designed that the output is as flat a
frequency response possible in the passband. The Butterworth filter is also called maximally flat
magnitude filter. The frequency response of the Butterworth filter has no ripples in the passband
and rolls off towards zero in the stopband [11]. It has slower roll-off and more linear phase
response when compared to other filters like Chebyshev and elliptic filter. Butterworth filters are
advantageously used to filter EEG signals as the pass- and stop-band are maximally flat, which
results in quality output signal for different frequency band.
The low-pass filter can be modified into high-pass filter; when placed in series with others to
form band-pass and band-stop filters. The gain G(ω) of an n-order Butterworth low pass filter
[12] in terms of transfer function H(s) is

17
( ) ( ) n
c
G
jHG 2
2
022
1 





+
==
ω
ω
ωω (1)
Where n is order of filter, ωc is cut-off frequency and G0 is the DC gain i.e. gain at zero
frequency.
The Butterworth filter is used to pre-process the EEG signal to remove low and high frequency
artefacts. The EEG signals are filtered with a second order Butterworth filter with cut-off
frequencies in a range of 5 - 30 Hz. In the second stage of filtering a moving average window
with Spencer’s 7 point is utilized. The moving average window [13] is weighted and given by
∑=
+=
q
r
rrt rnXay
0
)(
(2)
Where ar is a set of weights , Xt is the input time series
The symmetric weights for the spencer’s 7 point moving average window is given by [-0.05874
0.05874 0.29371 0.41257 0.29371 0.05874 -0.05874 ].
Once the filtering is done the maximum and the average energy is extracted in the frequency
domain using the Discrete Cosine Transform (DCT) [14]. DCT is a transformation method for
converting a time series signal into basic frequency components. Some simple functions to
compute the DCT and to pre-process the provided EEG data for BCI system are as follows:
The discrete cosine transform of a list of n real numbers s(x), x = 0, ..., n-1, is the list of length n
given by:
∑
−
=
+
=
1
0 2
)12(
cos)()(/2)(
n
x n
ux
xsuCnuS
π
u = 0,….,n (3)
Where C(u) is equal to 1/ square root of 2 for u=0 or is equal to 1 for all other values. Each
element of the transformed list S (u) is the inner product of the input list s(x) and a basis vector.
The constant factors are chosen so that the basis vectors are orthogonal and normalized.
The list s(x) can be recovered from its transform S(u) by applying the inverse cosine transform
(IDCT):
∑
−
=
+
=
1
0 2
)12(
cos)()(/2)(
n
u n
ux
uSuCnxs
π
x=0, …, n (4)
This equation expresses s as a linear combination of the basis vectors. The coefficients are the
elements of the transform S, which may be regarded as reflecting the amount of each frequency
present in the input s. Energies are obtained for all the channels individually using the above said
method.
The Naïve Bayes is a probabilistic method used for prediction[14]. During training the
conditional probabilities of each attribute in the predicted class is estimated from the training data
set. The parameter’s mean and variance obtained from a small set of training data is sufficient for

18
classification. Commonly used as the Naïve Bayes provide good results and easy interpretation of
the results[15,16].
The disadvantage being that the classifier considers that the occurrence of attributes is
independent; therefore the correlation between the attributes is ignored.
Instance based learner uses the Euclidean distance based algorithm to obtain the class of the test
instances[17]. Instance based learners use a weighted overlap of the feature values of test instance
and a memorized example. Metrics used in Instance based learners combine a per feature distance
metric with global feature weights that account for relative differences in discriminative power of
the features.
In this work we use two datasets available publicly to validate the proposed methodology. The
first dataset is obtained from experiments jointly conducted by University of Tübingen, Germany,
Dept. of Computer Engineering and Institute of Medical Psychology and Behavioural
Neurobiology, and Max-Planck-Institute for Biological Cybernetics, Tübingen, Germany, and
Universität Bonn, Germany, Dept. of Epileptology. In the experiment conducted the subject
imagines the movement of left small finger or tongue. The time series of the electrical brain
activity was picked up using 64 electrodes placed in the contralateral motor cortex. The second
data set is provided by Fraunhofer FIRST, Intelligent Data Analysis Group, and University
Medicine Berlin, Department of Neurology. In the second experiment 118 electrodes were used
with motor imageries of left hand and right foot. Figure I and figure II shows the time domain
signal of some electrodes during the motor activity.
Figure I : 3.5 second epoch from one of the electrodes
Figure II : 3.5 second signal during imagination of hand movement.

19
Figure III shows the output in the frequency domain of the input time domain data.
Figure III. The processed signal for feature extraction in frequency domain.
3. RESULT AND DISCUSSION
The classification accuracy using Naïve Bayes(NB) and Instance Based Learners(IBL) using the
proposed method is shown in figure IV and the relative absolute error is shown in figure V.
Figure IV : Classification accuracy of NB and IBL for the two dataset.
Figure V : Relative absolute error.

20
From figure IV and V it can be seen that Instance based learners are more consistent across
different datasets. Also being a simple mechanism for computing the class, IBL is proved to be
efficient and fast method for BCI.
4. CONCLUSION
In this paper the Brain Computer Interface classification problem was investigated. It was
proposed to extract maximum and average energy of EEG features using Discrete cosine
transform. As the frequency of interest is broadly in the region of alpha and beta waves,
Butterworth band pass filter was used to separate the frequency of interest. Moving average
Spencer window was used to remove noise among adjacent data values as EEG data is highly
noisy. Using 10 fold cross, the proposed feature extraction method was trained and tested using
Naïve Bayes classifier and Instance based learner. Two publicly available datasets were used to
test the proposed methodology. It is found that Instance based classifier provides consistent
results across different EEG datasets. Future work in the direction of using Neural Network
classifiers need to be investigated for improving the classification accuracy as the EEG data is
highly susceptible to noise.
REFERENCES
[1] N. Birbaumer, A. Kubler, N. Ghanayim, T. Hinterberger, J. Perelmouter, J. Kaiser, I. Iversen, B.
Kotchoubey, N. Neumann, and H. Flor. The Thought Translation Device (TTD) for completely
paralyzed patients. In IEEE Transactions on Rehabilitation Engineering, volume 8, pages 190–193,
2000.
[2] E. Donchin, K.M. Spencer, and R. Wijensinghe. The mental prosthesis: Assessing the speed of a
P300-based brain-computer interface. In IEEE Transactions on Rehabilitation Engineering, volume 8,
pages 174–179, 2000.
[3] R. Scherer, G.R. M¨uller, C. Neuper, B. Graimann, and G. Pfurtscheller. An asynchronously
controlled EEG-based virtual keyboard: improvement of the spelling rate. In IEEE Transactions on
Biomedical Engineering, volume 51(6), pages 979–984, 2004.
[4] D. J. McFarland, C. W. Anderson, K.-R. Muller, A. Schlogl, and D. J. Krusienski. Bci meeting 2005-
workshop on bci signal processing: feature extraction and translation. IEEE Transactions on Neural
Systems and Rehabilitation Engineering, 14(2):135 { 138, 2006.
[5] R. O. Duda, P. E. Hart, and D. G. Stork. Pattern Recognition, second edition.
WILEYINTERSCIENCE,2001.
[6] D. J. McFarland and J. R. Wolpaw. Sensorimotor rhythm-based brain computer interface (bci):
feature selection by regression improves performance. IEEE Transactions on Neural Systems and
Rehabilitation Engineering, 13(3):372-379, 2005.
[7] W. D. Penny, S. J. Roberts, E. A. Curran, and M. J. Stokes. Eeg-based communication: a pattern
recognition approach. IEEE Transactions on Rehabilitation Engeneering, 8(2):214-215, 2000.
[8] Bassani, T. and Nievola, J.C. “Brain-computer interface using wavelet transformation and naïve
Bayes classifier”,Advances in Experimental Medicine and Biology, Vol. 657, No. 2 , pp. 147-
165,2010.
[9] Rutkowski, T.M., Cichocki, A., Tanaka, T, Mandic, D.P., Jianting Cao, and Ralescu, A.L. “
Multichannel spectral pattern separation - An EEG processing application”, IEEE International
Conference on Acoustics, Speech and Signal Processing, Washington, DC, USA, 2009.
[10] Lotte, F., Lecuyer, A. and Arnaldi, B. “FuRIA: An Inverse Solution Based Feature Extraction
Algorithm Using Fuzzy Set Theory for Brain–Computer Interfaces”, IEEE Transactions on Signal
Processing , Vol.57, No.8, pp.3253-3263, 2009.
[11] Giovanni Bianchi and Roberto Sorrentino (2007). Electronic filter simulation & design. McGraw-Hill
Professional. pp. 17–20. ISBN 9780071494670.
[12] In Wireless Engineer (also called Experimental Wireless and the Wireless Engineer), vol. 7, 1930, pp.
536–541 - "On the Theory of Filter Amplifiers"-S. Butterworth

21
[13] Cadzow.J.A,Wilkes.D.M, Peters.R.A,Li.X.Image texture synthesis-by-analysis using moving-average
models. IEEE Transactions on Aerospace and Electronic Systems, Volume: 29 , Issue: 4, pp: 1110-
1122, 1993
[14] Singer.B, Veloso.M.M.Automating the modeling and optimization of the performance of signal
transforms. IEEE Transactions on Signal Processing, Volume : 50 , Issue:8 , pp: 2003-2014. 2002.
[15] Besserve.M,Garnero.L,Martinerie.J.Cross-Spectral Discriminant Analysis (CSDA) for the
classification of Brain Computer Interfaces. 3rd International IEEE/EMBS Conference on Neural
Engineering, 2007. CNE '07. pp:375 - 378,2007
[16] Coyle. D, McGinnity.T.M,Prasad.G.Identifying Local Ultrametricity of EEG Time Series for Feature
Extraction in a Brain-Computer Interface.29th Annual International Conference of the IEEE
Engineering in Medicine and Biology Society, 2007. pp:701-704.
[17] Cufoglu.A,Lohi.M,Madani.K.Classification accuracy performance of Naïve Bayesian (NB), Bayesian
Networks (BN), Lazy Learning of Bayesian Rules (LBR) and Instance-Based Learner (IB1) -
comparative study. International Conference on Computer Engineering & Systems, 2008. pp:210-
215.
Authors
Aparna Chaparala is working as an Associate Professor of the department of computer science and
engineering in R.V.R. & J.C. Collegeof Engineering, Chowdavaram,Guntur. She has 10 years of teaching
experience. She completed her M.Tech in Computer Sicence & Engineering. She is doing her research in
the area of Data Mining. Presently pursuing Ph.D from J.N.T.U, Hyderabad. She has published 4 papers in
international journals.
Dr J.V.R. Murthy is presently working as a professor in the department of CSE at J.N.T.U., Kakinada. He is
awarded Ph.D. in CSE from JNTU. He did M.Tech in CSE at IIT. He has 20 years of teaching experience
and 3 years industrial experience. A Memento of Appreciation was awarded for “good performance and on
schedule completion of People Soft HRMS project” by Key Span Energy Corporation, New York. He has
more than 15 publications in national and international journals. His interested areas of research include
Data Warehousing, data mining and VLDB.
Dr B. Raveendra Babu has obtained Masters degree in Computer Science and Engineering from Anna
University, Chennai. He received Ph.D. in Applied Mathematics from S.V University, Tirupati. He is
currently leading a Team as Director (Operations), M/s.Delta Technologies (P) Ltd.,Madhapur, Hyderabad.
He has 26 years of teaching experience. He has more than 25 international & national publications to his
credit. His interested areas of research include VLDB, Image Processing, Pattern analysis and Wavelets

ENERGY COMPUTATION FOR BCI USING DCT AND MOVING AVERAGE WINDOW FOR NOISE SMOOTHENING

More Related Content

What's hot (20)

Similar to ENERGY COMPUTATION FOR BCI USING DCT AND MOVING AVERAGE WINDOW FOR NOISE SMOOTHENING (20)

Recently uploaded (20)

ENERGY COMPUTATION FOR BCI USING DCT AND MOVING AVERAGE WINDOW FOR NOISE SMOOTHENING