See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/335569888

Design and Implementation of Efficient Digital Filter for Preprocessing of EEG

Signals

Conference Paper · March 2019

CITATIONS READS

0 1,622

3 authors:

Anshul Khatter Dipali Bansal

Gurugram University Manav Rachna International University
4 PUBLICATIONS   7 CITATIONS    64 PUBLICATIONS   326 CITATIONS   

SEE PROFILE SEE PROFILE

Rashima Mahajan
G D Goenka University, Gurgaon
18 PUBLICATIONS   114 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Identification of heart beat abnormality View project

All content following this page was uploaded by Anshul Khatter on 06 September 2019.

The user has requested enhancement of the downloaded file.

 Proceedings of the 13th INDIACom; INDIACom-2019; IEEE Conference ID: 46181
2019 6 International Conference on “Computing for Sustainable Global Development”, 13th - 15th March, 2019
th

Bharati Vidyapeeth's Institute of Computer Applications and Management (BVICAM), New Delhi (INDIA)

Design and Implementation of Efficient Digital Filter

for Preprocessing of EEG Signals
Anshul Dipali Bansal Rashima Mahajan
Department of ECE Department of ECE Department of CSE
Manav Rachna International Institute of Manav Rachna International Institute of Manav Rachna International Institute of
Research & Studies Research & Studies Research & Studies
Faridabad, INDIA Faridabad, INDIA Faridabad, INDIA
[email protected] [email protected] [email protected]

Abstract— In recent researches, electroencephalography Many techniques are applied by researchers for artifact
(EEG) gains widespread popularity in the physiological index removal in EEG signal like independent component analysis,
which permits incessant monitoring of the human’s workload in Recursive Least Squares adaptive filter, Spatially Constrained
an appropriate manner. EEG signals are delicate to cognitive load Independent Component Analysis(SCICA), average artifact
variations in numerous chores. The EEG signals acquired from subtraction(AAS), Blind Source Separation(BSS), and Wavelet
human scalp are contaminated with the distinct set of artifacts, Denoising [11], [12], [13], [14]. An Independent Component
most of the time. These artifacts affect the uniqueness of the Analysis (ICA) and Recursive Least Squares (RLS) are
signal due to which medical psychoanalysis and information employed to get rid of eye movement artifacts. A separate
retrieval may be difficult. Therefore, EEGs are first preprocessed
electrode is situated near the eyes for the excerpt of a reference
to eradicate the present artifacts to make these signals ready for
further signal processing. In this paper, digital infinite impulse
signal. This electrode records the horizontal and vertical
response and finite impulse response filter are implemented to
movements of eyes. Every reference input is envisaged into the
identify an efficient filter structure for preprocessing of acquired ICA field and after that the interference is assessed by the RLS
EEG signals. The performance analysis has been done by algorithm and then the assessed interference is deducted from
calculating the signal to noise ratio and cross-correlation. The the EEG signal in the ICA domain [12].
results show that finite impulse response filters outperform The artifacts are affected by the characterization of
infinite impulse response (IIR) filter while removing noise from spatiotemporal-frequency. An improved spatiotemporal-
the EEG signals. The focus of this research is on eliminating
frequency performance was proposed which provide a superior
random noise in EEG data, however this approach can be
extended to another source of EEG contamination. signal-to-noise ratio and clutter-to-signal ratio score. An
approach of template matching along with manifold tools of
Keywords— Cognitive workload; Electroencephalography; FIR signal-processing was assessed and verified on real EEG
Filter, IIR Filter, Preprocessing, noise signals. It has been observed that this method has enhanced
competency in EEG studies and analysis [15]. The
I. INTRODUCTION consequences of all types of artifact removal are combined and
Now a days, research on automated analysis of the EEG to their overall performances were 88.09% specificity and
assess mental states of human-like the cognitive workload is 89.438% sensitivity. But the deficiency of this research is that
progressively emerging day by day[1], [2]. Cognitive workload the recital of the algorithm was assessed on limited data sets
indicates the capacity of mental demand carried out through only. An amalgam of independent component analysis and
some task [3], [4]. EEG signals are classified on the basis of regression with high-order data was used for elimination of
frequency: alpha waves (frequency range 8 to 13 Hz), beta artifacts from EEG record [16]. The obtained results are
waves (frequency greater than 13 Hz), theta waves (frequency compared with another accessible method of ICA, wavelet-
range 3.5 to 7.5 Hz), delta waves ( frequency 3 Hz or less) [5], ICA, regression analysis and regression ICA and the analysis
[6]. While acquiring EEG signals, artifacts are also expose that mixing of ICA with regression conserves the
present. The unwanted noise generated during the recording of neuronal linked activity with EEG signals and eliminate ocular
the brain signals is known as artifacts. [7], [8]. In the artifacts. Other method does not dependent on a number of
encephalogram signal, the removals of artifacts are EEG channels used is carried out for each EEG channel
critical. Artifacts are of two kinds: Physiological and Non- locations. This method is universal as it does not depend on a
Physiological. The physiological artifacts occur due to the number of EEG channels used and this technique is not
movement of the head, blinking of an eye, rotation of eyeball, restricted to the artifact type which makes it competent to
sweating, heartbeat and contraction of muscle and non- apply online on diverse experiments and subjects [17]. To
physiological artifacts occur due to the external faults like a eradicate the ocular artifact zone, discrete wavelet transform
failure of the electrode, power supply and ventilation [9], [10] (DWT) with an adaptive method is used. The ocular artifact is

Copy Right © INDIACom-2019; ISSN 0973-7529; ISBN 978-93-80544-32-8 18

 Proceedings of the 13th INDIACom; INDIACom-2019; IEEE Conference ID: 46181
2019 6 International Conference on “Computing for Sustainable Global Development”, 13th - 15th March, 2019
th

majorly affected by the low-frequency and DWT is used as a preprocessing of EEG signals and identification of efficient
filter bank to classify low- frequency zone. The threshold filter structure.
procedure is also used to eradicate noise but the appropriate
selection of threshold value is essential to get the true EEG
signal [18]. Further, DWT based techniques may lead to an
extensive feature set.
As real EEGs are nonstationary in nature, the assessment of
a real signal is much complex than a simulated signal [19]. An
evolutionary Stone‟s blind source separation (ESBSS)
algorithms automatically take off the heartbeat and ocular Fig. 1. Block diagram of the filter structure identification system.
artifacts without the use of notch filter [20]. A data-driven
surrogate based artifact removal (SuBAR) is a propitious A. Loading of data
method for portable environments to cataloguing and to filter
the ocular/muscular artifacts from single-channel As literature findings indicate the elimination of artifact in
electroencephalography. In addition to portable environments, encephalogram is crucial and various techniques are used for
this technique is also proficient in the existence of slight and eliminating artifacts. In the proposed work, different filter
severe artifacts and provides much better efficiency than any structures are designed in MATLAB (R2016a, 64bit) to
other conventional single-channel EEG denoising technique identify the efficient filter structure for the elimination of
[21]. artifacts. The EEG signals can be load from online database
Physiobank ATM to Matlab workspace using „load‟ command
The blending of independent component analysis, discrete for further processing. EEG signal dataset in the desired format
wavelet transforms (DWT) and wavelet independent is introduced from the Neuroelectric and Myoelectric database
component analysis (WICA) is employed to remove of Physiobank ATM. In this database EEG signals from 11 fit
components of artifact [22]. A prior information of artifact is participators at frequency of 5, 6, and 10 Hz are acquired.
obtained online and puts them in wavelet independent These EEG signals were acquired during a rapid serial visual
component analysis (WICA) along with the data which presentation (RSVP) task using eight different channels of 10-
includes artifact. The components of artifact are acknowledged 20 EEG signal acquisition system (PO8, PO7, PO3, PO4, P7,
and detached by sorting the association of the noticeable a P8, O1, and O2). Here, „P‟ represents parietal region and „O‟
priori information of artifact and WICs [23]. represents occipital region. Four different EEG signals of
frequency 10Hz from the PO7 channel is considered in the
A digital filter is also used to eliminate the noise present in
proposed work. The loaded EEG signals are continuous signal
the EEG signal [24]. The digital filters are divided into two
and amplitudes are of the order of microvolt with the sampling
categories - Infinite Impulse Response (IIR) filter and Finite
rate of 2048 Hz [31], [32].
Impulse Response (FIR) filter [25], [26]. The impulse response
is of unlimited duration in the IIR filter while it is of narrow B. Preprocessing
duration in the FIR filter. The IIR and FIR filter is selected The preprocessing step involves the addition of random
according to the need and specifications of frequency response noise to the loaded EEG signals and implementation of digital
and afterward a suitable method is applied like frequency filter structures (IIR and FIR) to identify the best filter
sampling method, fourier series method or window method, structure. Noise is defined as a random signal that occurs
through which the most appropriate FIR filter structure will be naturally. There are numerous kinds of random noise like pink
obtained [27]. There are numerous advantages of the FIR filter noise, square root noise, white noise, blue noise, proportional
over IIR: phase-linearity, bounded-input bounded-output noise and thermal noise. Random noise occurs due to the
stability, sensitivity [28]. Because of these advantages, FIR random motion of electrons in any conducting medium.
filters are suitable for several applications [29], [30]. Whenever signals of random noise are combined with
The rest of the paper is structured as follows: Section II electronic circuits, the ensuing noise is equivalent to the
depicts the design of efficient digital filter. Section III collective power of individual signals. In the proposed work
describes implementation of the design filter. Section IV white noise is consisting of normally distributed random
describes the experimental results obtained with the detail numbers is used. The MATLAB function „randn‟ is used to
discussion. The final section gives a conclusion with the create a 1-by-N vector of random numbers with standard
possible future scope of this work. deviation of 0.1. The standard deviation is a way to calculate
how extreme the signal oscillates from the mean. There is no
II. DESIGN OF EFFICIENT DIGITAL FILTER precise peak to peak value in the random noise. The peak to
This paper includes the design set up to identify an efficient peak value in the random noise is around 6 to 8 times the
digital filter structure for artifact rejection from EEG signals. standard deviation so random noise of the peak to peak 0.6 to
A block diagram to identify the possible efficient filter 0.8 is generated in present research. A random noise generated
structure is sketched in Fig1. The proposed system design by randn function generates a sequence of numbers which have
consists of three main steps namely loading of data, not generated any definite pattern. This implies that the same
sequence of numbers cannot be reproduced [33].

Copy Right © INDIACom-2019; ISSN 0973-7529; ISBN 978-93-80544-32-8 19

Design and Implementation of Efficient Digital Filter for Preprocessing of EEG Signals

In order to filter out the noises present in EEG signal, C. Identification of efficient filter structure
different digital filter structures (IIR and FIR) are proposed. A This section proposed the parameters used to evaluate the
filter is used to reject unwanted frequencies from the input performance of implemented filter structures to reject artifacts
signal and permit the desired frequencies. The range of from EEG signals. The parameters selected are signal to noise
frequencies of a signal which allow to pass from the filter is ratio (SNR) and cross correlation (CC) to evaluate the
known as passband and the range of frequencies that does not performance of the proposed filters in this research. The filter
allow to pass from the filter is known as stopband. The which provides the higher signal to noise ratio and higher
following digital filters have been proposed to reject the added cross-correlation is considered as an appropriate filter for
noise from EEG signals. artifact removal from EEG signals. The signal to noise ratio is
1) IIR Filter defined as the ratio of signal power to the noise power and is
usually expressed in decibels. SNR is calculated in Matlab by
Infinite impulse response filter. Due to the feedback computing the ratio of the signal summed squared magnitude
present in the IIR filter, the impulse response of IIR filter is to that of the noise. The other parameter cross correlation
infinite. IIR digital filters are represented by equation 1 as determines the correlation between the original and
follows: reconstructed (filtered) EEG signal. It is calculated by finding
∑ the coefficient of the correlation between the original and
H(z) = ∑
(1) reconstructed signal.
where Bk is the feedforward filter coefficients, Ak is the The loading of EEG signals to MATLAB workspace,
feedback filter coefficients and K is the order of the filter. preprocessing of EEG signals and identification of more
efficient filter structure can be summed up in the form of an
2) FIR Filter efficient MATLAB algorithm, as depicted in Figure 2.
Finite impulse response filter. In the FIR filter, the impulse
response is finite because there is no feedback in the filter. FIR
digital filters are represented by equation 2 as follows:
H(z) = ∑ (2)
where Bk is the feed forward filter coefficients, M is the length
of the filter and K is the order of the filter[34]
Bandstop IIR filter of the second order with lower cutoff
frequency 50 Hz and higher cutoff frequency 60Hz having the
sampling rate of 512 Hz is proposed to remove the noise from
EEG signal. The bandstop filter is a selective filter with a high
rejection which cannot attenuate other frequencies that belongs
to the EEG signal. The IIR filters of the second order are
usually known as biquads which are used to implement higher
order filters by cascading biquad. The order of a filter is
relatively proportional to the number of calculations involved
or the number of components required to design a filter.
Therefore, by selecting a low order filter, the computation time
can be minimized. A high-order filter offers a better response;
but the high-order filter is costly, occupies more space and
offers more complexity in design. That‟s why a proper
selection of order is necessary for designing a filter.
Fig. 2. Detailed work flow to identify an efficient filter structure to reject
A low pass filter can be used to eliminate noise at higher artifacts from acquired EEG signals
frequencies than that of EEG signals. A low pass IIR filter
with cutoff frequency of 50Hz, the passband ripple of 0.01 III. IMPLEMENTATION OF DESIGN FILTER
with a sampling rate of 512 Hz is proposed to remove the noise The EEG signals are loaded from online database
from EEG signal. Further, a low pass FIR filter with a cut off Physiobank ATM to Matlab workspace using „load‟ command
frequency 50Hz and the sampling rate of 512 Hz is proposed for further processing. The four different EEG signals of
by varying the order of the filter to identify at which range of frequency 10Hz from the PO7 channel is considered in the
the order FIR filter gives a better response. The phase delays research work. A white random noise consisting of normally
introduced by the FIR filters can be nullified by applying zero- distributed random numbers is added to the loaded EEG signal.
phase digital filtering using Matlab function filtfilt ( ). In zero- The implementation of different filter structures to identify the
phase digital filtering the EEG data is processed in both the efficient filter structure for the elimination of artifacts is done
directions (forward and reverse) which results in the zero in MATLAB (R2016a, 64bit). In order to filter out the noises
phase distortion. present in EEG signal, different digital filter structures (IIR

Copy Right © INDIACom-2019; ISSN 0973-7529; ISBN 978-93-80544-32-8 20

 Proceedings of the 13th INDIACom; INDIACom-2019; IEEE Conference ID: 46181
2019 6 International Conference on “Computing for Sustainable Global Development”, 13th - 15th March, 2019
th

and FIR) have been implemented. The bandstop IIR filter of A random noise is added to loaded EEG signal records.
the second order with lower cutoff frequency 50 Hz and higher The original EEG signal, noise signal and resulted noisy EEG
cutoff frequency 60Hz having the sampling rate of 512 Hz is signal is shown in Figure4.
implemented to remove the noise from the EEG signal. Original signal
e 100
d
A low pass filter is used to eliminate the noise at higher u
itlp
m
0

frequencies than that of EEG signals. A low pass IIR filter A

-100
0 1 2 3 4 5 6 7 8 9 10
with cutoff frequency of 50Hz, the passband ripple of 0.01 Time [s]
Noise
with a sampling rate of 512 Hz is implemented to remove the e
d
0.5

noise from EEG signal. The Matlab function „designfilt‟ is u

itl
p
m
0

used to design digital filters. Further, a low pass FIR filter with A
-0.5
0 1 2 3 4 5 6 7 8 9 10
a cut off frequency 50Hz and the sampling rate of 512 Hz is Time [s]
Original signal + Noise
implemented by varying the order of the filter to identify at e 100
d
u
which range of the order FIR filter gives a better response. The ti
l
p
m
0

phase delays introduced by the FIR filters are nullified by A

-100
0 1 2 3 4 5 6 7 8 9 10
applying zero-phase digital filtering using Matlab function Time [s]

filtfilt( ). Fig. 4. An original EEG signal, random noise signal and resulted noisy EEG
signal
The parameters are used to evaluate the performance of
implemented filter structures to reject artifacts from EEG Figure 4 shows an original EEG signal loaded from
signals. The parameters selected are signal to noise ratio physionet and then random noise is introduced in it. After that,
(SNR) and cross correlation (CC) to evaluate the performance it shows the original signal with random noise. The bandstop
of designed filters in this research. The filter which provides IIR filter of second order with lower cutoff frequency 50 Hz
the higher signal to noise ratio and higher cross-correlation is and higher cutoff frequency 60Hz having a sampling rate of
considered as an appropriate filter for artifact removal from 512 Hz is implemented to eliminate the noise from the EEG
EEG signals. SNR is calculated in Matlab by computing the signal and the filtered signal obtained is shown in Figure 5.
ratio of the signal summed squared magnitude to that of the
noise. The other parameter cross correlation determines the
correlation between the original and reconstructed (filtered)
EEG signal. It is calculated by finding the coefficient of the
correlation between the original and reconstructed signal.
IV. RESULTS AND DISCUSSION
This section describes and analyzes the results of
experimentation. The proposed algorithm to identify the
efficient filter structure for artifact removal from EEG signals
was developed. The algorithm for the proposed approach is
developed in MATLAB release (R2016a) on the core i2
processor. In this experiment, the developed system is tested
and performance has been evaluated using EEG signals loaded
from the database of Physiobank ATM. The EEG record is
exported to MATLAB workspace from the database of
Physiobank ATM and is shown in Figure 3. Fig. 5. Filtered EEG signal using bandstop IIR filter

A low pass IIR filter with the cutoff frequency of 50Hz, the
passband ripple of 0.01 with a sampling rate of 512 Hz is
implemented to remove the noise from EEG signal. The
Matlab function „designfilt‟ is used to design digital filters and
the filtered signal is obtained is as shown in Figure 6.
The results shows that low pass IIR filter gives the best
results when the filter order is 8 and pass band ripple is very
less (approximately equal to zero). Then to evaluate the
performance of the filter, performance parameters (SNR and
CC) are calculated initially when no filter is implemented and
then after implementing the filter.
Table I shows that low pass IIR filter outperforms the band
stop IIR filter while rejecting noise artifacts from EEG signals.
The analysis of performance parameter results do not show any
Fig. 3. An EEG signal loaded from the database of Physiobank ATM significant improvement in SNR values even with low pass IIR

Copy Right © INDIACom-2019; ISSN 0973-7529; ISBN 978-93-80544-32-8 21

Design and Implementation of Efficient Digital Filter for Preprocessing of EEG Signals

filter structures. This justifies the need of further exploring the the SNR and CC increases but after the order 5 SNR and CC
filter structures possessing high SNRs and CCs. A low pass values start decreasing with the increase in order of the filter.
FIR filter with a cut off frequency 50Hz having the sampling This shows for signal2, fir low pass filter with order 5 provides
rate of 512 Hz is implemented by varying the order of the an improved result. The trends of signal 3 and 4 are same as
filter. The phase delays introduced by the FIR filter is nullified signal1 and the comparison of data obtained reveals that the
by applying zero-phase digital filtering using Matlab function best performance is obtained corresponding to maximum SNR
filtfilt( ). A FIR low pass filter is applied and the filtered signal and CC with FIR low pass filter possessing order 4 and 5
is obtained as shown in Figure 7. respectively.

(a)

Fig. 6. Filtered EEG signal using low pass IIR filter

TABLE I. PERFORMANCE PARAMETERS OF NOISY EEG SIGNAL AND

FILTERED EEG SIGNAL USING B ANDSTOP IIR FILTER AND L OW PASS IIR
FILTER

Filters applied SNR CC

No filter 43.180489 db 0.999976
Bandstop IIR Filter 43.189671 db 0.999976
Low Pass IIR Filter 43.326977 db 0.999977

(b)
Fig. 8. Performance analysis of IIR and FIR filters for EEG signals using (a)
SNR, and (b) CC

The performance analysis of implemented IIR and FIR

filter structures for EEG signal preprocessing is depicted in
Fig8a using SNR values and in Fig 8b using CC values,
respectively. Similar trends of results are obtained from other
Fig. 7. Filtered EEG signal using low pass FIR filter signals also after filtering. The analysis reveals that the
maximum value of SNR and CC is obtained when an order4
The performance parameters are calculated by applying FIR filter is applied to an input noisy EEG signal. An effort is
filters to four different EEG signals and by varying the order of made to develop a more efficient and robust filter to eliminate
low pass FIR filter. The value of SNR and CC of the first noisy noise artifacts from acquired EEG signals which can be used to
EEG signal is found to be 43.0577db and .999975 which classify different levels of cognitive workload using MATLAB
improves on implementation of FIR filter. As the order of the based algorithms. A key disadvantage of the FIR filter is while
filter increases, the value of SNR and CC improves however implementing the filter, huge arithmetic operations are
after the order4 SNR and CC values start decreasing with involved which confines its speed and more power is required.
further increase in order of filter. This shows the FIR low pass
filter with order4 provides an improved result for the first V. CONCLUSION
signal. With signal 2, the value of SNR and CC of the noisy In this paper, an accurate and efficient digital filter
signal is 44.7118db and .999983 which improves when the FIR implementation algorithm to eliminate noise/artifacts from
filter is applied. As the order of the filter increases the value of EEG signals has been presented. The performance parameters

Copy Right © INDIACom-2019; ISSN 0973-7529; ISBN 978-93-80544-32-8 22

 Proceedings of the 13th INDIACom; INDIACom-2019; IEEE Conference ID: 46181
2019 6 International Conference on “Computing for Sustainable Global Development”, 13th - 15th March, 2019
th

(signal to noise ratio and cross correlation) have been of this research is to apply designed filter structures on real-
calculated to assess the recital of designed filter structures. The time EEG signals. It shall be tested while adding different
results show that a FIR filter with order 4 to 5 possesses better types of noises to real time EEG signals during the
filtering capacity with higher SNR and CC than the IIR filter in preprocessing phase. These preprocessed EEG signals will
denoising different EEG signals loaded from the Physiobank further be used for the identification of the associated cognitive
ATM database. A more efficient filter structure is identified to workload by developing efficient signal processing algorithms.
remove noise from input noisy EEG signals. The future scope
TABLE II. PERFORMANCE PARAMETERS OF FOUR NOISY EEG SIGNALS AND FILTERED EEG SIGNALS FILTERED USING LOW PASS FIR FILTER

FIR filter Signal1 Signal2 Signal3 Signal4

applied SNR (in db) CC SNR (in db) CC SNR (in db) CC SNR(in db) CC
No filter 43.0577 .999975 44.7118 .999983 43.6896 .999978 36.1873 .999880
Order 1 47.2647 .999991 48.7638 .999993 47.9265 .999992 40.4464 .999955
Order 3 47.5465 .999992 49.0778 .999994 48.0354 .999993 40.6212 .999957
Order 4 47.7004 .999993 49.0512 .999994 48.3778 .999993 40.8384 .999959
Order 5 47.5382 .999991 49.1731 .999994 48.1265 .999992 40.6624 .999957
Order 6 47.1680 .999990 48.2995 .999993 47.7469 .999992 40.3032 .999953
Order 8 46.6847 .999989 48.0464 .999993 47.2754 .999991 39.6846 .999946

REFERENCES [12] C. Mosquera and A.Vazquez, “Automatic removal of ocular artifacts

from EEG data using adaptive filtering and Independent Component
[1] S. Chandra, K. Verma, G. Sharma, A. Mittal and D. Jha, “EEG based Analysis,” in Proc. of the 17th IEEE European Signal Processing
cognitive workload classification during NASA MATB-II multitasking,” Conference, 2009, pp.2317-2321.
International Journal of Cognitive Research in Science, Engineering and
[13] Z. Wang, P. Xu, T. Liu, Y. Tian, X. Lei, and D. Yao, “Robust removal of
Education, vol. 3, no. 1, pp. 35-42, 2015.
ocular artifacts by combining Independent Component Analysis and
[2] M. Jong-Hagelstein, W. Sandt-Koenderman, N. Prins, D. Dippel, P. system identification,” Biomedical Signal Processing and Control, vol.
Koudstaal, and E. Visch-Brink, “Efficacy of early cognitive-linguistic 10, pp. 250-259, 2014.
treatment and communicative treatment in aphasia after stroke: a
[14] S. Goh, H. Abbass, K. Tan, A. Al-Mamun, C. Wang, and C. Guan,
randomised controlled trial (RATS-2),”, Journal of Neurology,
Neurosurgery & Psychiatry, vol. 82, no. 4, pp. 399-404, 2010. “Automatic EEG Artifact Removal Techniques by Detecting Influential
Independent Components,” IEEE Transactions on Emerging Topics in
[3] B. Cain, “A review of the mental workload literature,” Defence Research Computational Intelligence, vol. 1, no. 4, pp. 270-279, 2017.
and Development, Canada, Toronto., 2007.
[15] A. Eldeen, A. Farag, and A. Allah, “A Hybrid Approach for Artifacts
[4] J. Li, Y. Wang, L. Zhang, A. Cichocki, and T. Jung, “Decoding EEG in Removal from EEG Recordings,” International Journal of Computer
Cognitive Tasks With Time-Frequency and Connectivity Masks,” IEEE Applications, vol. 168, no. 4, pp. 10-19, 2017.
Transactions on Cognitive and Developmental Systems, vol. 8, no. 4, pp.
298-308, 2016. [16] M. Mannan, M. Jeong, and M. Kamran, “Hybrid ICA—Regression:
Automatic Identification and Removal of Ocular Artifacts from
[5] T. Aich, “Absent posterior alpha rhythm: An indirect indicator of seizure Electroencephalographic Signals,” Frontiers in Human Neuroscience,
disorder?,” Indian Journal of Psychiatry, vol. 56, no. 1, pp. 61-66, 2014. vol. 10, pp. 1-17, 2016.
[6] T. Johannisson, “Correlations between personality traits and specific [17] I. Winkler, S. Haufe, and M. Tangermann, “Automatic Classification of
groups of alpha waves in the human EEG,” PeerJ, vol. 4, 2016. Artifactual ICA-Components for Artifact Removal in EEG Signals,”
[7] T. Radüntz, J. Scouten, O. Hochmuth, and B. Meffert, “EEG artifact Behavioral and Brain Functions, vol. 7, no. 1, pp. 1-30, 2011.
elimination by extraction of ICA-component features using image [18] V. D. Sanaja and P. Patel, “A combination of RLS algorithm with
processing algorithms,” Journal of Neuroscience Methods, vol. 243, pp. Wavelet transform for Detection and Removal of OA from EEG signal,”
84-93, 2015. American International Journal of Research in Science, Technology,
[8] A. Khatter, “EEG Ocular Artifact Elimination by Extraction of ICA Engineering & Mathematics, vol. 10, no.2, pp. 161-165, 2015.
Component,” International Journal of Scientific Research in Science, [19] S. Valipour, G. Kulkarni, and A. Shaligram, “Study on Performance
Engineering and Technology, vol. 3, no. 6, pp. 893-895, 2017. Metrics for Consideration of Efficiency of the Ocular Artifact Removal
[9] P. Bhuvaneswari and J. Satheesh Kumar, “Methods used for Identifying Algorithms for EEG Signals,” Indian Journal of Science and
EEG Signal Artifacts,” in Proc. of the International Conference on Technology, vol. 8, no. 30, 2015.
Computational Intelligence and Information Technology, 2012, pp. 375- [20] A. Abdullah, C. Zhang, A. Abdullah, and S. Lian, “Automatic Extraction
379. System for Common Artifacts in EEG Signals based on Evolutionary
[10] A. Khatter, D. Bansal, and R. Mahajan, “Study of Various Automatic Stone‟s BSS Algorithm,” Mathematical Problems in Engineering, vol.
EEG Artifact Removal Techniques,” International Journal for Research 2014, pp. 1-25, 2014.
in Applied Science & Engineering Technology, vol. 5, no.10, pp.1027- [21] M. Chavez, F. Grosselin, A. Bussalb, F. De Vico Fallani, and X.
1037, 2017. Navarro-Sune, “Surrogate-Based Artifact Removal From Single-Channel
[11] S. Khatun, R. Mahajan, and B. Morshed, “Comparative analysis of EEG,” IEEE Transactions on Neural Systems and Rehabilitation
wavelet based approaches for reliable removal of ocular artifacts from Engineering, vol. 26, no. 3, pp. 540-550, 2018.
single channel EEG,” in Proc. of the IEEE International Conference on [22] C. Zhang et al., “Automatic Artifact Removal from
Electro/Information Technology (EIT),2015, pp.335-340. Electroencephalogram Data Based on A Priori Artifact Information,”
BioMed Research International, vol. 2015, pp. 1-8, 2015.

Copy Right © INDIACom-2019; ISSN 0973-7529; ISBN 978-93-80544-32-8 23

Design and Implementation of Efficient Digital Filter for Preprocessing of EEG Signals

[23] A. Bigirimana, N. Siddique, and D. Coyle, “A hybrid ICA-wavelet [28] A. Chandra and S. Chattopadhyay, “Design of hardware efficient FIR
transform for automated artefact removal in EEG-based emotion filter: A review of the state-of-the-art approaches,” Engineering Science
recognition,” in Proc. of the IEEE International Conference on Systems, and Technology, an International Journal, vol. 19, no. 1, pp. 212-226,
Man, and Cybernetics (SMC), 2016. 2016.
[24] P. Podder, T. Zaman Khan, M. Haque Khan, and M. Muktadir Rahman, [29] L.Tan, Digital Signal Processing: Fundamentals and Applications.
“Comparative Performance Analysis of Hamming, Hanning and Academic Press, New York, 2011.
Blackman Window,” International Journal of Computer Applications, [30] S. Zahoor and S. Naseem, „Design and implementation of an efficient
vol. 96, no. 18, pp. 1-7, 2014. FIR digital filter,” Cogent Engineering, vol. 4, no. 1, 2017.
[25] N. E. Diana, U. Kalsum, A. Sabiq, W. Jatmiko, and P. Mursanto, [31] A. Goldberger et al., “PhysioBank, PhysioToolkit, and PhysioNet,”
“Comparing Windowing Methods on Finite Impulse Response (FIR) Circulation, vol. 101, no. 23, 2000.
Filter Algorithm in Electroencephalography (EEG) Data Processing,”
Journal of Theoretical and Applied Information Technology, vol. 88, [32] A. Matran-Fernandez and R. Poli, “Towards the automated localisation
of targets in rapid image-sifting by collaborative brain-computer
no.3, pp. 558-67, 2016.
interfaces,” PLOS ONE, vol. 12, no. 5, 2017.
[26] J. Proakis, Digital Signal Processing. Pearson, 2014.
[33] S. Smith, The scientist and engineer's guide to digital signal processing.
[27] R. Mehboob, S. Khan, and R. Qamar, “FIR filter design methodology for San Diego, Calif.: California Technical Pub., 1997, pp. 14-29.
hardware optimized implementation,” IEEE Transactions on Consumer
[34] S. Salivahanan, A. Vallavaraj and C. Gnanapriya, Digital signal
Electronics, vol. 55, no. 3, pp. 1669-1673, 2009.
processing. New Delhi: Tata McGraw Hill Education Pvt. Ltd., 2010, pp.
539-554.

Copy Right © INDIACom-2019; ISSN 0973-7529; ISBN 978-93-80544-32-8 24

View publication stats