Content uploaded by Chowdhury Azimul Haque
Author content
All content in this area was uploaded by Chowdhury Azimul Haque on Jul 20, 2019
Content may be subject to copyright.
2019 International Conference on Electrical, Computer and Communication Engineering (ECCE), 7-9 February, 2019
978-1-5386-9111-3/19/$31.00 ©2019 IEEE
An Approach to Estimate the Activation of Different Bands
of EEG Signal using Classified Songs
Chowdhury Azimul Haque, Monira Islam, Abdul Munem Saad and Md. Salah Uddin Yusuf
Department of Electrical and Electronic Engineering
Khulna University of Engineering & Technology
Khulna-9203, Bangladesh
Email: c.azimulhaque@gmail.com, monira_kuet08@yahoo.com, munemsaad@yahoo.com, ymdsalahu2@gmail.com
Abstract— Human cognition analysis from EEG signal is one
of the most significant way to find the causes which affect human
brain. Songs are familiar stimuli to analyze the effectiveness of
different EEG bands because it is a common source of
entertainment. According to linguistic variation, subject age and
preference, volume level of songs, the impact on different EEG
bands varies. In this research, a work has been performed to
estimate the effect of songs on different EEG bands (delta, theta,
alpha, beta). The songs have been categorized into three types:
mild, pop, rock according to their PSD value which are 0.2114
W/Hz, 0.3271 W/Hz and 0.8448 W/Hz respectively. The volume
levels are considered as 0 %-14% volume level for low volume,
26%- 60% volume level for comfortable volume and 61%-
100% volume level for high volume. An ANOVA test has been
conducted to indicate the significance variation in EEG band
and in which P-value has been set 0.05. For the mild, pop, and
rock songs the P-values for alpha bands are 0.0146, 0.006 and
0.0006 respectively and the values indicate that alpha band is
mostly active while listening to different songs. A topographical
representation has been done based on the effects of songs on the
alpha band activity for mild, pop and rock songs at three volume
level. The maximum percentage of alpha band activation is 60%
in comfortable volume which indicates that, when the volume is
high human cognition state moves from relax to stress condition
in which beta band is active.
Keywords—Electroencephalogram (EEG), Songs, Analysis of
variance (ANOVA), Topography.
I. INTRODUCTION
Electroencephalogram or EEG is a record of the electric
signal generated by action of brain cells. In other word EEG
can be defined as voltage fluctuation in the brain due to ionic
current flow within neurons [1]. EEG is an effective
procedure to record the brain electrical activity directly [2].
EEG signals can be classified into five wavebands i.e., delta,
theta, alpha, beta, gamma bands according to their signal
frequency. Some classification can be done with the shape
and symmetry property of the signal [3]. Different bands of
EEG are affected due to various condition of the brain and the
effect of the condition in the brain can be defined by the
characteristic of the EEG bands [4]. Table I shows the band
frequency, amplitude and corresponding human condition.
Delta band ranges from 0.1 Hz to 4 Hz and grey matter of
brain is represented by this band. Theta band ranges from 4
Hz to 8 Hz and this type of band helps to increase relaxation
and get relief from pain. Alpha band ranges from 8 Hz to 13
Hz and it is the most important band to analysis the brain
effect. This band represents awake but relaxed state with
closed eyes. The range where beta band can be found is 13
Hz to 30 Hz. Beta bands are associated with active and busy
state of the brain. Gemma band ranges from 30 Hz to 70 Hz.
These waves are activated for hyper alertness and both the
senses and memories are combined in these bands [5].
TABLE I. FREQUENCY BAND AND CONDITION OF HUMAN TO
CORRESPONDING BAND
Brain Waves
Frequency band
Condition
Delta (Δ)
1-4 Hz
Deep Sleep
Theta (Ɵ)
4-8 Hz
Light Sleep
Alpha (α)
8-13 Hz
Relaxation period
Beta (β)
13-30 Hz
Active period
Gamma (ɤ)
30- 40 Hz
Abnormal condition
Nowadays many research works are done showing that
different types of music have different effects on the brain.
EEG responses of human brain during listening to different
music is studied by Independent Component Analysis (ICA)
in [6]. The level of enjoyment during music listening is
studied by calculating cross-correlation between sound
stimuli and EEG signals and a topographical analysis is
presented in [7].The effect of music on central nervous
system (CNS) from EEG signal using average mutual
information and phase-space reconstruction of the signal is
experimented in [8].EEG band variation during musical
stimuli has beendetermined by band symmetry, sample
entropy and statistical analysis i.e., ANOVA test in [9].
Authors in [10] had investigated the signatures of EEGby
calculating power spectral density of EEG bandsfor two
different music phrases. It has beendescribed in [11] that
while listening to passive unconstrained music the frequency
bands of EEG is increased in power level. Many studies have
claimed that music stimuli can activate alpha band or beta
band. In [12], authors had showed that a relax music
stimulican activate alpha band by reducing the beta band and
it is analyzed by independent component analysis.In this
research work, the effect of three types of music: mild, pop,
rock which are categorized based on their power spectra has
studied. These three types of music have enormous effect on
EEG bands specially on alpha and beta band. These bands
effects are analyzed with statistical and topographical
approach. The effect of volume level of the songs on the EEG
band is also studied in this work. Song categorization method
is shown in Fig. 1(a) andthe entire research work is described
in Fig.1(b).
The rest of the paper is organized as follows: Section II
describes the methodology used in this work, Section III
narrates the result and discussion. Finally, Section IV
concludes the result.
II. METHODOLOGY
A. Subjects and Data Acquisition
Eight healthy volunteers who had no chronic and hearing
disorder have been participated in this study. The partici-
pants were verbally informed and provided written consent
(a)
(b)
Fig. 1. Proposed workflow diagram of (a) music categorization, (b) Effect
of songs on EEG bands
prior to participation in this study. It is also confirmed by the
author that data have been recorded without doing any
violation to Helsinki Principle [13].
9 channel B-Alert X-10 wireless system has been used in
this study for data acquisition. This system acquires EEG data
from three vital region of the brain i.e., frontal, central and
parietal lobe. According to the International 10/20 system of
Electrode placement on the scalp for recording EEG data
[10/20 system], this system covers the F3, Fz, F4, C3, Cz,C4,
P3, Poz, and P4 positions. The sampling rate of the EEG data
that can be recorded by B-Alert X10 device is 256 per second
and dynamic range is ±1000 µV which is sufficient for EEG
[14].
In this study, thirty songs have been taken as data
acquisition stimuli. According to the Power spectral Density
value of the songs, they have been categorized into three major
type: Mild, Pop and Rock. Each type of song has three trials
and they were of averagely 60 seconds long with sample rate
of 44,100 Hz. Three trials of three types of songs were used as
stimuli for every participant and these songs were played to
the participants with three different volume level i.e., low
volume, comfortable volume and high volume through
earphones. As control EEG signal of relaxed state with eyes
closed was recorded for each participant before data
acquisition for the concerned stimuli. Participants were sitting
on a fixed chair with the eyes closed during data acquisition
so that any artifact related with eye and body movement could
be neglected in the EEG signals. It has taken seven minutes
for each data acquisition for every song with three different
volume level. Total data acquisition protocol can be defined
by the diagram shown in Fig. 2.
B. Artifact and noise removal of EEG data
Although EEG data represent the activity of brain, it is also
contaminated with electrical activities arising from other sites.
These contaminated electrical activities are called artifact
which can be divided into two types: (i) Physiological Artifact
and (ii) Extra-physiological Artifact. Physiological artifacts
are related with body movement, eye blinking and Extra-
physiological artifacts are related with equipment and
environment. The main artifact which can be considered as the
problem for data processing in this study is the noise which is
created by electrode popping [15].
Fig. 2. Proposed approach for data acquisition.
(a)
(b)
(c)
Fig. 3. (a) Raw, (b) Notch filtered, (c) Elliptic filtered EEG data.
To remove this type of artifacts in recorded EEG data, a
tunable notch filter has been designed. A narrow-band filter
has been considered as notch filter. The amplitude response
Hd(s) of a notch filter can be expressed as Eq. (1) [16].
(1)
Where, given notch frequencies
satisfy ≤ for
1≤ p ≤ q which is allowed to distribute in the range [0, π].
k0andk1 of Eq. (1) can be defined as following:
k0=
(2)
Where, is rejection bandwidth.
k1= [ 0, π] - k0 (3)
In this research, notch filter is used to eliminate the power
line noise in 50 Hz. The raw data, notch filtered data and
elliptic filtered data is shown in Fig.3. During designing notch
filter, the tunable frequency was considered as 48 Hz and 52
Hz to keep the notch frequency in the range of 50 Hz.
C. EEG Band Extraction
EEG contains five bands i.e., delta, theta, alpha, beta,
gamma band which corresponding frequencies are given in
Table. I. It is important to separate the bands from EEG data
for extracting the both frequency and time domain features.
In this research, frequency-domain i.e., spectral analysis is
done. For extracting the features from the recorded EEG data
elliptic filter is used. Elliptic filter is also referred to as Cauer
filter. An elliptic filter can be designed by adding zeros to
poles. This type of filter has shorter transition as it allows
ripple in both the stop band and pass band [17]. The
magnitude squared frequency response of an elliptic filter,
can be represented as Eq. (4)
(4)
Where, is elliptic rational function of angular
frequency determined from the specified ripple factor .
A large value of indicates the large transition band and
small value indicates the small one. In this work, the bands of
EEG (Delta, Theta, Alpha, Beta, Gamma) has been extracted
by varying the pass band and the stop band of the designed
elliptic filter. The gain of the pass band can be varied between
1 and
. And for the stop band this is between 0 and
; where is the discrimination factor which is
defined with the rational function . The bands have
been extracted through setting the pass band and stop band
according to the frequency range of the respective band. The
ripple factor was set to 0.1 decibel and the attenuation was 30
decibels. The extracted bands from the recorded EEG signal
is shown in Fig. 4.
D. Feature Extraction
Frequency-domain features are connected to extract
different signals including bio-signals. For analyzing EEG
signal, it is necessary to convert the EEG signal from time-
domain to frequency-domain. In this research, spectral
analysis of the EEG signal has been done using Fast Fourier
Transform (FFT) algorithm to calculate the power spectral
density (PSD). PSD indicates the ranges of the frequencies at
which variations are strong and weak. PSD calculation is non-
parametric and Welch method is the most prominent method
for EEG analysis18]. The PSD can be defined as the Fourier
transform which is expressed in Eq. (5).
∞
∞ (5)
Where, rx(k) means the auto correlation for the periodic signal
and represents the PSD. But for non-periodic
process the can be written as Eq. (6).
Fig. 4. Extracted EEG bands (delta, theta, alpha and beta) from recorded
raw signal.
−=
+=
+
→
N
Nn
nxknkr N
N
x)()(
lim
)( 12
1
(6)
Where, denotesthe analysis of convolution.
If it is assumed that successive sequences are offset by
Dpoints and that each sequence are L point long, then the ith
sequence of the signal can be expressed as Eq. (7)
(7)
Where,denotes the signal sequence.
According to conditions for nonparametric signal
Welch’s method can be written as data record which is shown
in Eq. (8)
(8)
Where, U data points are covered entirely by K sequences and
the overlap is L-D point and denotes the amplitude
response.
Logarithmic power calculates the overall power of a
signal [19]. This power is considered as regional power of the
brain in this study. To calculate the logarithm of the power, it
is necessary to calculate the power of each EEG band.
(9)
Where,
nq
x
is the magnitude of qth frequency band of the nth
sample and
B
P
is the power in a specified frequency band.
So that, log power, can be written as,
(10)
III. RESULT AND DISCUSSION
A. Song Categorization for Hearing Stimuli
For hearing stimuli, the selected songs have been
categorized in three major types: Mild, Pop, and Rock based
on their spectral analysis i.e., power spectral density values.
From their spectral analysis, difference between their genre
can be identified. Their average PSD values and correlation of
their signal has been shown in Table II. and Fig.5 respectively.
Fig. 5, shows that the correlation among the songs are less,
therefore, a significant difference between the categorized
songs can be defined. These songs have been listened by the
participants in three different volume: low (0%-14% volume
level), comfortable (15%-60% volume level) and high (61%-
100% volume level).
B. Effective band selection with statistical analysis
A two-way ANOVA (Analysis of variance) [20] was
conducted to find the significant variations of EEG signal
among the bands with respect to the comfortable and high-
volume level. ANOVA is a way to find out the significance of
the experiment results by dividing the values into two groups.
Here one group variable was the lobes (Frontal, Central, and
Parietal) and the other group was prepared with the
corresponding log power values of the regarding stimuli. The
same analysis was performed on the PSD values of the EEG
signal. The significance level was set to 0.05 and if the P-value
of the test greater than the significance level there is no
variation in the results. From the results, we found that both
the values of log power and PSD had the significant variations
in the three lobes. The variations were found for the alpha and
beta bands. When the participants were listening to mild songs
in medium volume level, the variation in alpha band (F (2,8)
=3.9, 5.02; P-values< 0.05) has been found. The variation in
beta band (F (2,8) = 5.13, 6.07); P-values< 0.05) has been also
found in the experiment. These variations denote that, there is
a difference between the effects when a subject listens the
mild songs in low and in comfortable volume. In addition,
there is a significant variation for both pop and rock songs in
lobe to lobe and volume level to volume level difference.
TABLE II. THE CATEGORIES OF THE SONGS ACCORDING TO THEIR MEAN
POWER SPECTRAL DENSITY VALUE
Song Type
Mean PSD (Watts per Hertz)
Mild
0.2114
Pop
0.3272
Rock
0.8448
Fig. 5. Correlation among mild, pop and rock songs.
Table III shows the detail information of F-values and P-
values of the corresponding band for log power and PSD
values respectively. Table IV shows the property variations
among the lobe with respect to the low and high-volume level
in which other group of two-way ANOVAs is corresponding
PSD values. From Table IV it is found that for the mild songs
alpha band (F (2,8) = 7.06, 9.75; P-values< 0.05) shows the
TABLE III. ANOVA TEST RESULT FOR LOG POWER FOR EEG BANDS
(VARIATION IN LOBE TO LOBE)
Song
Type
Bands
F values (2,8)
P values
Alpha
F (2,8) =3.9, 5.02
0.0416, 0.003
Mild
Beta
F (2,8) = 5.13, 6.07
0.0189, 0.0011
Delta
F (2,8) = 0.79, 2.24
0.4696, 0.0816
Theta
F (2,8) = 1.7, 6.71
0.2134, 0.0006
Alpha
F (2,8) = 8.91, 14.59
0.006, 0.0003
Pop
Beta
F (2,8) = 4.68, 2.46
0.0368, 0.1054
Delta
F (2,8) =2.49, 2.56
0.1328, 0.0963
Theta
F (2,8) = 3.08, 33.02
0.0907, 0
Alpha
F (2,8) = 16.82, 17.26
0.0006, 0.0001
Rock
Beta
F (2,8) = 2.49, 1.11
0.1323, 0.4139
Delta
F (2,8) = 0.48, 2.79
0.6332, 0.0784
Theta
F (2,8) = 1.32, 13.48
0.3097, 0.0004
TABLE IV. ANOVA TEST RESULT FOR POWER SPECTRAL DENSITY OF
EEG BANDS
Song
Type
Bands
F values (2,8)
P values
Alpha
F (2,8) =7.06, 9.75
0.0122, 0.0013
Mild
Beta
F (2,8) = 9.31, 20.7
0.0052, 0.0001
Delta
F (2,8) = 1.02, 2.31
0.3964, 0.1214
Theta
F (2,8) = 0.64, 1.97
0.5474, 0.1684
Alpha
F (2,8) = 13.57, 1.3
0.0014, 0.3368
Pop
Beta
F (2,8) = 17.64, 5.24
0.0005, 0.0128
Delta
F (2,8) =1.53, 5.71
0.2631, 0.0096
Theta
F (2,8) = 12.22, 10.07
0.0021, 0.0012
Alpha
F (2,8) =23.32, 4.3
0.0002, 0.0239
Rock
Beta
F (2,8) = 5.16, 2.04
0.0289, 0.1577
Delta
F (2,8) = 0.01, 3.16
0.9865, 0.0390
Theta
F (2,8) = 5.98, 4.17
0.0196, 0.0262
variation for both lobe to lobe difference and volume level to
level difference. The F-value denotes the interaction term
which is basically used for establishing relationship between
factor and significance level. From the entire discussion for
the ANOVA test result it can be stated that, there is a variation
between in EEG bands while listening to different songs at
different volume.
C. Brain Mapping for Different EEG Band
The resultant brain mapping for the mild songs at
comfortable volume level (15%-25%) the alpha band of the
recorded EEG signal which has been collected with F3, F4,
P3 and P4 positioned electrode is activated. At medium
volume level (26%-60%), activation of alpha band cannot be
significantly described. And for the high-volume level (61%-
100%) the activation of alpha band is negligible compared to
the other volume level. The activation of alpha band for mild
songs at different volume level is shown in Fig. 6. For the pop
songs at comfortable volume level the alpha band is activated
in P3 and P4 position of the brain. At low volume level the
activation of alpha band throughout the brain less than that of
comfortable volume level. At high-volume level, alpha band
activity for the pop songs is merely identified in the
corresponding topography. Fig. 7 shows the alpha band
activity of the EEG signal for pop songs at different volume
levels. Similar process is performed for estimating the alpha
band activity for rock songs through topographical
representation. Fig. 8(a) shows that, for the rock songs at
comfortable volume level the alpha band activity is
dominating throughout the brain. At low and high-volume
level, the alpha is less dominating for the rock songs.
It has also showed that, the activity of alpha band for mild
songs at comfortable volume in brain is greater than the pop
and rock songs. Thepercentage of alpha band activation
throughout the brain for different songs at different volume
level is described in Table. V.
From these topographical analyses it can be stated that, the
alpha band of EEG signal is the most active band for these
three types of songs. But with the increasing volume level, the
activity of alpha band is decreasing which indicates the
increasing activity of beta band in the for these songs. The
increasing beta band activity attests the stress condition of
human brain.
TABLE V. PERCENTAGE OF ALPHA BAND ACTIVITY IN BRAIN FOR THE
SONGS AT DFFERENT VOULME LEVEL
Fig. 6. Brain mapping through topographical representation for
estimating alpha band activity of mild song at (a) comfortable volume, (b)
low, (c) high volume.
Fig. 7. Brain mapping through topographical representation estimating
alpha band activity of pop song at (a) comfortable volume, (b) low volume,
(c) high volume.
Fig. 8. Brain mapping through topographical representationestimating
alpha band activity of rock song at (a) comfortable volume, (b) low volume,
(c) high volume.
IV. CONCLUSION
This experiment evaluates the effect of mild, rock and pop
songs on different bands of EEG signal which have been
categorized according to their power spectral density. The
correlation coefficient varies from 0.0318 to 0.048 which
indicates that there exists largest variation among different
types of songs which affects differently on individual EEG
bands. Tunable notch filter is being used for signal filtering
which is passed through an elliptic filter for extraction of
individual EEG bands. A two-way ANOVA was designed to
investigate significant variation among the bands. Brain-
mapping has been performed for the topographical
representation. While listening songs at comfortable volume
level (15%-60%), largest variation is observed for alpha band
which is 60%, 52%, 45% for mild, pop and rock songs
respectively which keeps the subject relax while for lower or
highest volume level the activation region of alpha band
decreases and beta band gets more activated which keeps the
subject more stressed. So, this is an innovative approach to
indicated the activity of different EEG bands while listening
songs which will play a significant role in physiological
research area of human cognition.
REFERENCES
[1] J. Suto and S. Oniga, “Music Stimuli Recognition in
Electroencephalogram Signal,” Elektronika ir Elektrotechnika, vol. 24,
no. 4, 2018.J. Clerk Maxwell, A Treatise on Electricity and Magnetism,
3rd ed., vol. 2. Oxford: Clarendon, 1892, pp.68–73.
[2] K. Zeng, J. Yan, Y. Wang, A. Sik, G. Ouyang, and X. Li, “Automatic
detection of absence seizures with compressive sensing
EEG,” Neurocomputing, vol. 171, pp. 497–502, 2016.
[3] R. Salmelin and R. Hari, “Spatiotemporal characteristics of
sensorimotor naromagnetic rhythms related to thumb movement,”
Neuroscience, vol. 60, no.2, pp. 537-550, 1994.
[4] C. R. Hema, M. p Paluraj, s. yaccob, and R. Nagarajan, “An Analysis
of the Effect of EEG Frequency Bands on the Classification of Motor
Imagery Signals,” Biomedical Soft Computing and Human Sciences,
vol. 16, no. 1, pp. 121–126, Dec. 2009.
[5] J. S. Kumar and P. Bhuvaneswari, “Analysis of
Electroencephalography (EEG) Signals and Its Categorization–A
Study,” Procedia Engineering, vol. 38, pp. 2525–2536, 2012.
[6] W.-C. Lin, H.-W. Chiu, and C.-Y. Hsu, “Discovering EEG Signals
Response to Musical Signal Stimuli by Time-frequency analysis and
Independent Component Analysis,” 2005 IEEE Engineering in
Medicine and Biology 27th Annual Conference, 2005.
[7] Y. Kumagai, M. Arvaneh, and T. Tanaka, “Familiarity Affects
Entrainment of EEG in Music Listening,” Frontiers in Human
Neuroscience, vol. 11, 2017.
[8] A. Dey, S. K. Palit, D. K. Bhattacharya, D. Tibarewala, and D. Sarkar,
“Study of the effect of different music stimuli on autonomic nervous
system of a single subject,” 2014 International Conference on
Communication and Signal Processing, 2014.
[9] R. Nawaz, H. Nisar, and Y. V. Voon, “The Effect of Music on Human
Brain; Frequency Domain and Time Series Analysis Using
Electroencephalogram,” IEEE Access, vol. 6, pp. 45191–45205, 2018.
[10] R. S. Schaefer, R. J. Vlek, and P. Desain, “Music perception and
imagery in EEG: Alpha band effects of task and
stimulus,” International Journal of Psychophysiology, vol. 82, no. 3,
pp. 254–259, 2011.
[11] A. Markovic, J. Kühnis, and L. Jäncke, “Task Context Influences Brain
Activation during Music Listening,” Frontiers in Human
Neuroscience, vol. 11, 2017.
[12] V. Straticiuc, I. E. Nicolae, R. Strungaru, T. M. Vasile, O. A. Bajenaru,
and G. M. Ungureanu, “A preliminary study on the effects of music on
human brainwaves,” 2016 8th International Conference on Electronics,
Computers and Artificial Intelligence (ECAI), 2016.
[13] “WMA - The World Medical Association-WMA Declaration of
Helsinki – Ethical Principles for Medical Research Involving Human
Subjects,” The World Medical Association. [Online]. Available:
https://www.wma.net/policies-post/wma-declaration-of-helsinki
Song
Type
Volume level
Low
(0%-14%)
Comfortable
(15%-60%)
High
(61%-100%)
Mild
35%
60%
25%
Pop
30%
52%
23%
Rock
30%
45%
15%
ethical-principles-for-medical-research-involving-human-subjects/.
[Accessed: 27-Oct-2018].
[14] B-Alert X-10 user manual. Biopac systems, Inc., 2014.
[15] “EEG Artifacts,” Overview, Physiologic Artifacts, Extraphysiologic
Artifacts, 02-Jul-2018. [Online]. Available:
https://emedicine.medscape.com/article/1140247-overview.
[Accessed: 27-Oct-2018].
[16] W. Xu, A. Li, B. Shi, and J. Zhao, “A Novel Design of Sparse FIR
Multiple Notch Filters with Tunable Notch
Frequencies,” Mathematical Problems in Engineering, vol. 2018, pp.
1–7, 2018.
[17] R. Schaumann and V. V. M. E., Design of analog filters. Oxford:
Oxford University Press, 2010.
[18] F. Khanam, M. A. Rahman, and M. Ahmad, “Evaluating alpha relative
power of EEG signal during psychophysiological activities in salat,”
International Conference on Innovations in Science, Engineering and
Technology (ICISET), 27-28 October, IUC, Chittagong, Bangladesh,
2018.
[19] T. H. Aspiras and V. K. Asari, “Log power representation of EEG
spectral bands for the recognition of emotional states of mind,” 2011
8th International Conference on Information, Communications &
Signal Processing, 2011.
[20] S. Boslaugh, “Introduction to Regression and ANOVA,” in Statistics
in a Nutshell, O'Reilly Media, Inc., 2013.
[21] L. S. Hooi, H. Nisar, and Y. V. Voon, “Tracking of EEG activity using
topographic maps,” 2015 IEEE International Conference on Signal and
Image Processing Applications (ICSIPA), 2015.
