EAI Endorsed Transactions

on Pervasive Health and Technology Research Article

Elementary Time Frequency

Analysis of EEG Signal Processing
1,* 1 1
Parag Puranik , Dr. R.V. Kshirsagar and Sagar Motdhare
1
PIGCE, Nagpur

Abstract
EEG analysis is used as a tool for medical diagnosis of various diseases related to brain like epilepsy, dementia, brain
disorders like stress etc. To identify such neural disorders it is essential to process these signal because of the non-
stationary behavior. The time frequency analysis of EEG signals provide the correct visualization of EEG signals to
extract the various rhythms of frequencies like alpha, beta, gamma waves. The Wavelet transform is used to carry out the
time frequency analysis of EEG signal. Wavelets are used to analyze the EEG signals correctly. Their applications
include seizure analysis, modeling of neuron potentials etc. To perform these applications it is necessary to perform the
mathematical operations on EEG signal by windowing, filtering and then finally reconstruction.

Keywords: EEG, Morlett Window, Wavelets, montages, Wavelet Transforms.

Received on 01 December 2017, accepted on 11 May 2018,published on 23 July 2018

Copyright © 2018 Parag Puranik ET AL., licensed to EAI. This is an open access article distributed under the terms of the
Creative Commons Attribution licence (http://creativecommons.org/licenses/by/3.0/), which permits unlimited use,
distribution and reproduction in any medium so long as the original work is properly cited.

doi: 10.4108/eai.13-7-2018.155081

performed out using discrete Fourier Transform (DFT),

discrete cosine transform (DCT) etc. But these tools for
1. Introduction analysis has an disadvantage that spectrum of the signals is
degraded due to short term windowing analysis and fixed
EEG is a tool used to study the brain activities. The transforms. Parametric estimation methods of EEG signals
need to develop the EEG as a tool is used for detecting the like Auto-regression (AR) models have an advantage over
various brain disorders like stress, epilepsy, dementia etc. DCT of correct representation of frequency domain analysis
EEGs are use to identify the neural activities of the brain. but has disadvantage of improper estimation of model
These neural activities are diagnosed by multiple-electrode parameters since the measured signal is of very limited
EEG machines which are placed according to the standard length. EEG signals are statistically non-stationary signals.
montages. The EEG signals are time domain signals but Because there are many abnormal events occurring while
these signals provide other information related to brain in capturing EEG signals. [1]
frequency domain with imaging tools to visualize
topographies. The time frequency analysis of EEG signal is EEG signals are statistically non-stationary signals.
performed using Wavelet transforms to visualize the signals Because there are many abnormal events occurring while
correctly. Modeling and localization of EEG signals is capturing EEG signals the analysis of EEG signals become
important to obtain the correct information about these important. During this analysis frequency domain
signals after capturing through EEG electrodes. The exact parameters are incorporated over the specified interval and
monitoring of patients is required for real time applications do not calculate the individuality of the signal correctly.
provided by EEG sequences. The major information is Therefore, time frequency analysis is needed. In short, to
extracted by the sleeping brain and therefore sleep EEG is analyze the EEG and extract the correct frequency
carried out. [10]. components and rhythms time frequency analysis is worked
out. [1][2][9]
EEGs are basically non-stationary signals. The analysis of
stationary signals is easy and can be easily represented in time The paper is divided as follows: Section –II depicts the
or frequency domain. The frequency domain analysis is basic EEG and placement of electrodes as per the

*
Corresponding author. Email: [email protected]

EAI Endorsed Transactions

1 on Pervasive Health and
Technology
02 2018 - 07 2018 | Volume 4 | Issue 14 | e4
Parag Puranik, Dr.R.V. Kshirsagar, Sagar Motdhare

international standard. Section – III examines the Wavelet

transform and need for time frequency analysis of EEG
signals. Section – IV describes the results obtained and
Section – V represents the conclusion.

2. Basics of EEG Recordings

2.1. EEG Recording

Figure 3. (a) 10-20 electrode placement system (b) 3
dimensional side view and top view
EEG analysis and recording procedure for patients is shown
below in figure 1. EEG is done on 24 channel RMS EEG
machine with 10-20 electrode system. The figure 2 shows
the placement of electrode. 24-recording electrodes 2.1. EEG Rhythms
including ground and system reference for EEG recording
for patients are used; these electrodes were located
according to 10–20 electrode system as follows: Fp1, Fp2, To treat with EEG signals and to extract the correct and
F7, F3, Fz, F4, F8, A1, T3, C3, Cz, C4, T4, A2, T5, P3, Pz, useful information from the EEG diagnosis, the rhythmic
P4, T6, O1, and O2. The 10-20 electrode placement system activities of EEG is to be considered. Clinical EEG wave
and its three dimensional top view and side view is shown forms amplitude that is typically between 10 and 100 v and
in figure.3. at a frequency range of 1Hz to 100Hz. EEG is classified
into the following five rhythms according to their frequency
(1) Alpha () wave: this regular wave is present in healthy
adults while they are conscious comfortable, and eyes closed.
It arises at a frequency range of 8Hz to 13Hz with a normal
voltage range of around 20 v to 200 v. waveform is mostly
pragmatic in the posterior region of thebands.
head.
(2) Beta ( ) wave: the frequency of waves ranges from
13Hz to
30Hz, and voltage range is 5 v to 10 v. waves

central nervous
appear with additional excitation of the
system, boost with awareness and alertness. waves are

observed in the parietal and frontal region of thescalp.

(3) Theta ( ) wave: the frequency range of wave is
Figure 1.EEG Recording Procedure 4Hz to 7Hz.
This waveform is important during sleep,

stimulation in older children and adults, exciting stress,

and idling. wave is recorded across the temporal and
parietal of the scalp with an amplitude range of 5 v to 10 v.

region as focusing,
This wave has been linked to activities such
awareness, psychological effort, and stimulation
processing.[2].
(4) Delta ( ) wave: the lowest frequency of wave is
3.5Hz, and its voltage ranges from 20 v to 200 v.

less than
wave occurs during deep snooze, in
newborn, and with

serious organic brain diseases. This waveform can be

recorded frontally in adults and posteriorly in children.
(5) Gamma( )wave: the frequency of wave ranges from
30Hz to 100Hz. This waveform is recorded in the
somatosensory cortex in the case of cross model sensory
processing, during short-term memory to distinguish
Figure 2.EEG Recording Procedure matter, sounds, perceptible impression, and in pathological
case because of cognitive decline.[2].
The recorded EEG needs successive stages of processing the
signal to extract important indications from the EEG signal of
stress patients, and these indications reflect brain medical

EAI Endorsed Transactions

2 on Pervasive Health and
Technology
02 2018 - 07 2018 | Volume 4 | Issue 14 | e4
 Elementary Time Frequency Analysis of EEG Signal Processing

6 -µ5 .27'2
3!! 4:

8/01
changes. The main stages of EEG signal processing are
denoising, feature extraction, and classification.
5 399 (5)

i. EEG recording: EEG signals are acquired

ii. EEG pre-processing: EEG signals are pre- EEG signal segmentation is carried out since this is useful
elaborated in order to improve the noise-signal for clinicians and neurologists. The segmented EEG signals
rate. are stationary signals with time frequency statistics. EEG
iii. Features extraction: relevant features are extracted, signals can be divided into three segments, posterior,
in order to detect specific waveforms (Time- occipital, a postictal segment which is highly beneficial to
Frequency analysis) obtain the various rhythmic frequencies of EEGs.[1][2]

iv. EEG classification: subjects’ EEG signals are While segmentation of EEG signals, the time frequency
characteristics of the signals can be oppressed. This causes
identified as belonging to appropriate classes the variation measurements between the adjoining EEG
[8][10]. frames. This is calculated based on autocorrelations for
segment of length N defined as: [1]
1 Dµ2µE
A '/ B= B '/ B=, 0,@ "1 I
3. Time-Frequency Analysis of EEG
signals ; , = >@ FGH
@,@B1
0

As stated, EEG signals are non-stationary. To extract the (6)

correct information from these signals it is necessary to
perform the analysis. To perform the analysis EEG is The windowing sequence using Hanning window is
considered as stationary signals for very short obtained for EEG segmented signal and auto-regressive
durations/intervals. EEG signal allocation is considered as based modeling of the signals is performed. The
Gaussian curve which calculates the mean and covariance prediction
for each segmented waveform. EEG signals are seen during error of the AR model [1] is defined as:
the sleep, wakefulness, during eye blinking i.e. where the LM=' (7)
alpha rhythms are strong. There is a change in distribution ",=
of the signal and this can be measured by the Gaussian ,='
process. The non-Gaussian parameters can be estimated by 3
skewness, kurtosis, negentropy and Kulback-Laiber (KL) . = " ∑MG2
distance. [1]. Where p is the forecast order and , k=1,2…p are aim is to minimize the

the prediction coefficients. The main LM=

Skewness is a evaluation of equilibrium of error between two samples/segments and windows of the
the EEG signal.
distribution and is defined as follows for real signals
The segmented central spikes of the EEG signal is
shown below in figure 4. The segmented EEG shows the
frontal electrodes data, frontal and central and posterior and
occipital montage.
¤µ§ (1)

Where µ and σ are mean and standard deviation

respectively
Kurtosis is measure to detect whether the data is
sharp or even relative to the normal distribution. The
kurtosis is
defined for normal distribution as:
"3
-!!¤

$%&¤' -¤ (2)

The negentropy is defined as :

( ¤')*+,, (3)
" (¤'

covariance ')*+,, is a Gaussian random signal with

Figure 4. Segmented central spikes of
Where
as x(n) and H(.) is differential entropy defined The short time Fourier transform (STFT) over a sliding window [1] is performed as follows:

as: (¤' 5
-µ5 .¤' 3¤
2 4' ∑RGµ55 µPQR (8)
(4)
/01 ' ON
is always .¤' is the signal distribution and negentropy Where n O
of the window. The
to the position
, N "
Where refers
non-negative. spectrogram is defined as: |T
, N|U (9)
The KL distance between two distributions p1 and p2 is
defined as:
,N
The windowing function is chosen properly for the
estimation of power spectrum. For correct time resolution
 EAI Endorsed Transactions
3 on Pervasive Health and
Technology
02 2018 - 07 2018 | Volume 4 | Issue 14 | e4
Parag Puranik, Dr.R.V. Kshirsagar, Sagar Motdhare

Hanning window of various durations of 1 and 2 msec gUQ

duration is obtained.[1]. qN f gQ
hY |N| < N2jI (17)
3.1. Wavelet Transform 0 hY Nj ≤ |N| < U

The important parts and coefficients can be better focused 4. Implementation

the signal components of EEG in time-frequency gap. This The following flowchart shows the implementation of time
is required to develop the need between time and frequency
components. Therefore, the main aim of wavelet transform frequency analysis using Wavelet Transform for EEG
by Morlet window is to have clarity time relative to the signals.
sampling period of EEG. The continuous wavelet transforms The EEG signal is recorded as shown in figure 1 and 2. The
are analyzed for this purpose. The Morlet continuous EEGs obtained contain lots of artifacts. Also EEGs are son-
wavelet transforms
VL, W
is given
√*
as: -µ5 YZ ∗ \ * ^ 4
stationary signals. Therefore, to extract correct information
Hanning window is used. Also Haar transforms are used to
2 5 ]µ* (10) from these EEGs signals and to carry out time frequency
analysis windowing of the signals is carried out. For this
is the analyzing vector, a(>0) is the scale parameter calculate the statistical analysis. Time frequency analysis is
b is the position parameter.
and Z carried out using wavelet transforms using Morlet window.
This version is linear and is invariant under translational This is done for spectrogram calculations. Then wavelet
and dilations. Morlet’s wavelet is a complex waveform transforms are analyzed using Fourier transforms. And lastly
defined as Z 2 µ]/UaPU_bc] (11) original EEG is reconstructed. This is done for all samples
of EEG.[1].
The reconstruction algorithm works as follows at low

√U_
resolution:[1][2]

The above wavelet is divided into real and imaginary

parts of sine and cos function; where b0 is constant. i. Compute the fast Fourier transform of the signal
3.1. Wavelet Transform using Fourier ii. Set ‘j’ i.e. data resolution and Compute the FFT of
the wavelet coefficients at scale ‘j’(Equations
Transform Z" (14) and (15)
continuous wavelets. If is d {Y c
iii. Obtain the product of wavelet coefficients Wj by
Consider the scalar product for G. (G=filter)
band limited to half of the iv. Calculate the Inverse Fourier transform considering
data can be properly sampled. The
sampling frequency, the Z
data at the decision j=1 is as follows: Wj and G
(12) v.Then perform j=j-1 and proceed for step3.
2

d2 d2 ¤ Y. U
Z U ]
"
time can be computed from dc with discrete
And
filter with the frequency response H(ώ).
gUQ (13) The continuous wavelet transforms (CWT) is used to
hY |N| < Nj
I
gQ

(N f 5. Results
2 analyze how the frequency content of a signal changes over

Therefore 0 hY Nj ≤ |N | < Unext scale can be time. (CWT) is used to analyze signals jointly in time and
the coefficients at the
calculated as follows: P frequency.
lma2N l PN(2 N (14) The figure 6 shows the wavelet transform and inverse
Wavelet transform of the EEG obtained from patient
The cut-off frequency is reduced by twice at each
level [1], thereby allowing to reduce the number of Figure 7 shows the wavelet transform using Morlet
samples. The window and its histogram analysis which provide the
Wavelet coefficients at the scale j+1 are statistical data of mean, standars deviation, median etc for
Pa2G Y, 2
µPa2
Z2
µPa2
" (15) all sampled values of EEG signals. The EEG signal of 1200
msec duration is analyzed. This signal is sampled and
And can be computed from Cj as follows: windowed for each 200 msec duration. And then its
P (16) continuous wavelet transform is calculated. The original
VPa2N l PNq2 N signal for each sample is reconstructed using the stated
algorithm. This provides the time frequency analysis and
Where G is the discrete time filter: wavelet coefficients.

4
on Pervasiv e Health and Technolo gy
EAI Endorsed Transactions

02 2018 - 07 2018 | Volume 4 | Issue 14 | e4

 Elementary Time Frequency Analysis of EEG Signal Processing

Acquisition of EEG Signals

using electrode placement

Windowing of EEG signal for

short duration

Time frequency analysis using

Wavelet transform

Calculation of spectrogram

Figure 6. Wavelet and Inverse Wavelet Transform of

Wavelet transform using EEG Signal
Fourier tansform

Reconstruction of the original

signal.

Any other
samples to be
calculated

Original EEG signal

(Reconstructed) Figure 7. Wavelet Transforms and Histogram
Analysis with statistical analysis

Figure 5. Flowchart of Implementation 6. Conclusion

In this paper, the basic concept of EEG signal processing
which include EEG signal segmentation, signal transform,
time frequency analysis using Wavelets is covered. The
wavelet transform and reconstruction algorithm is based on

5 on Pervasiv e Health and Technolo gy

EAI Endorsed Transactions

02 2018 - 07 2018 | Volume 4 | Issue 14 | e4

Parag Puranik, Dr.R.V. Kshirsagar, Sagar Motdhare

certain requirements which are used for specific Acknowledgements.

applications like seizure analysis, stress management etc. The authors would like to thank Saeid Sanei and J,A,
These are the future scope of the work wherein the EEG Chambers the authors of the book “EEG signal Processing” for
signal processing will be carried out. The time frequency providing the idea of the implementation of this paper. Further
analysis provides the statistical data of EEG signals which authors are thankful to Dr. Karandikar, MD, neurologist for
will be beneficial to classify the EEG signal. providing the medical background help.

References

[1] Saeid Sanei, J.A. Chambers, “EEG Signal Processing”,

Center of Digital Signal Processing, John Wiley and Sons
Ltd.
[2] Bassel Abou-Khalil, Karl Misulis, “Atlas of Textbook of
EEG & Seizure Semiology”, Textbook of Elsevier.
[3] Amjed S. Al-Fahoum and Ausilah A. Al-Fraihat, “Methods
of EEG Signal Features Extraction Using Linear Analysis
in Frequency and Time-Frequency Domains”, Hindawi
Publishing Corporation ISRN Neuroscience Volume 2014,
Article ID 730218.
[4] Davide V. Moretti, “Neurophysiology in Alzheimer’s
Disease and Dementia”, EDITORIAL published:
27June2016, Frontiers in Aging Neuroscience.
[5] Alida A. Gouw1,2 and Cornelis J. Stam,
“Electroencephalography in the Differential Diagnosis of
Dementia”, Epileptologie 2016, pp.-173-186.
[6] Anthoula Tsolaki, Dimitrios Kazis, Ioannis Kompatsiaris,
“Electroencephalogram and Alzheimer’s Disease: Clinical
and Research Approaches”, Hindawi Publishing
Corporation International Journal of Alzheimer’s Disease”,
Volume 2014, Article ID 349249
[7] Ales Prochazk, Jaromir Kukal, “ Wavelet Transform used
for feature extraction and EEG signal Segments
classification”, Institute of chemical technology, Prgaue,
http://dsp.vscht.cz, 2012
[8] Neep Hazarika, Jean Zhu Chen, “ Classification of EEG
signals using the Wavelet Transform”, Elsevier Signal
Processing Journal, Issue-1, Volume 59, May 1997, pp.-61-
72.
[9] Costanza D. Avanzo, Vincenza Tarantino, Patrizia
Bisiacchi,” A Wavelet Methodology for EEG time-
frequency Analysis in a Time Discrimination Task”,
International Journal of Bioelectromagnetism, Vol.11,
No.4, pp.-185-188, 2009.
[10] A.H. Siddiqi, H. Kodal Sevindir, C. Yazici, “Spectral
Analysis of EEG signals by using wavelet and Harmonic
Transforms”.

EAI Endorsed Transactions

6 on Pervasive Health and
Technology
02 2018 - 07 2018 | Volume 4 | Issue 14 | e4