See discussions, stats, and author profiles for this publication at: [Link]

net/publication/11352413

Comparison of Wavelet Transform and FFT Methods in the

Analysis of EEG Signals

Article in Journal of Medical Systems · July 2002

DOI: 10.1023/A:1015075101937 · Source: PubMed

CITATIONS READS
279 6,880

1 author:

Mehmet Akin
Dicle University
48 PUBLICATIONS 1,877 CITATIONS

SEE PROFILE

All content following this page was uploaded by Mehmet Akin on 05 December 2017.

The user has requested enhancement of the downloaded file.

P1: FJT/GAP P2: GYK/GAQ QC: FpQ
Journal of Medical Systems [joms] PP405-368576 March 23, 2002 11:32 Style file version Nov. 19th, 1999

Journal of Medical Systems, Vol. 26, No. 3, June 2002 (°

C 2002)

Comparison of Wavelet Transform and FFT Methods

in the Analysis of EEG Signals
M. Akin1

In this study, whether the wavelet transform method is better for spectral analysis of
the brain signals is investigated. For this purpose, as a spectral analysis tool, wavelet
transform is compared with fast Fourier transform (FFT) applied to the electroen-
cephalograms (EEG), which have been used in the previous studies. In addition, the
time-domain characteristics of the wavelet transform are also detected. The comparison
results show that the wavelet transform method is better in detecting brain diseases.
KEY WORDS: wavelet; FFT; EEG.

INTRODUCTION

Brain is one of the most important organ of the body. So, its structure and
functions have become a great source for the research. Although a lot of work has
already been done in this field, it is still a challenging area. Furthermore, EEG is an
important guide for the investigation of the physiological and psychological sutiation
of participants.(1,2)
EEG signals are considered not to be deterministic and they have no special
characterestics like ECG signals. In addition, when the Fourier transform is applied
to successive segments of an EEG signal, the obtained spectra are observed to be time
varying. This indicates that the EEG signal is also nonstationary. The spectral analysis
based on the Fourier transform classical method assumes the signal to be stationary,
and ignores any time-varying spectral content of the signal within a window.(3,4)
Furthermore many doctors, as a rule, determine the frequency of EEG signals
by counting the number of peaks per second. From this point of view, it can be said
that both frequncy and time-domain characteristics of EEG are very important.
In this study, EEG signals taken from patients in Neurology Clinic of Research
Hospital of Dicle University have been used. These EEG signals were digitized by
using a data processing unit (PCI-MIO-16-E4) which has been used in the previous

1 Department of Electric and Electronic Engineering of Dicle University, 21280 Diyarbakir, Turkey.

241
0148-5598/02/0600-0241/0 °
C 2002 Plenum Publishing Corporation
P1: FJT/GAP P2: GYK/GAQ QC: FpQ
Journal of Medical Systems [joms] PP405-368576 March 23, 2002 11:32 Style file version Nov. 19th, 1999

242 Akin

studies.(2) Digitized EEG signals were analyzed using fast Fourier transform (FFT)
and wavelet transform (WT) methods. The results obtained by the two techniques
were eventually compared, and it was found that WT offers better spectral charac-
teristics of EEG signals in the case of diagnosis.

MATERIALS AND METHODS

Characteristics of the EEG Waveforms

An EEG signal consists of several frequency bands, which are called δ, θ , α,
and β bands. Their bandwidths are 0–4, 4–8, 8–12, and above 12 Hz respectively. In
this study, two EEG signals for both healthy (normal) and pathological (abnormal)
cases were recorded from subjects under relaxation, with their eyes closed. The
EEG patterns recorded from subjects under relaxation show the characteristics of α
waveform, which is the case for most humans.(5) This condition seems to represent a
form of syncronization, almost like a natural frequency of brain.
The abnormal case is shown in Fig. 1. If one counts the number of peaks in
the trace, as physicians do, it is clear that low-frequency band is dominant in the
waveform. This low frequency indicates an abnormal situation to be taken under
consideration.
On the other hand, Fig. 2 shows the normal case, and it can be observed that the
dominant frequencies of EEG waveform are in the α range.

Fig. 1. Pathological EEG waveform.

P1: FJT/GAP P2: GYK/GAQ QC: FpQ
Journal of Medical Systems [joms] PP405-368576 March 23, 2002 11:32 Style file version Nov. 19th, 1999

Comparison of Wavelet Transform and FFT Methods in the Analysis of EEG Signals 243

Fig. 2. Healthy EEG waveform.

Signal Analysis Methods

The basic approach to signal analysis is to get proper information from the signal
by applying the best suitable method. The conditions of reversibility of the applied
method and representing the original signal by the converted form should be satisfied
one by one.(6) The methods used in this study are Fourier and wavelet transforms.

Fourier Transform
Spectral analysis of a signal involves decomposition of the signal into its fre-
quency (sinusoidal) components. In other words, the original signal can be separated
into its subspectral components by using spectral analysis methods.(7)
Among spectral analysis techniques, Fourier transform is considered to be the
best transformation between time and frequency domains because of it being time-
shift invariant. The Fourier transform pairs are expressed as(6)
X
N−1
X(k) = x(n)WNkn (1)
n=0

1 X
N−1
x(n) = X(k)WN−kn (2)
N k=0

where WN = e− j(2π/N) and N = length [x(n)].

The Fourier transforms of the EEG signals shown in Figs. 1 and 2 are displayed
as Figs. 3 and 4.
P1: FJT/GAP P2: GYK/GAQ QC: FpQ
Journal of Medical Systems [joms] PP405-368576 March 23, 2002 11:32 Style file version Nov. 19th, 1999

244 Akin

Fig. 3. Spectrum of the pathological waveform.

Fig. 4. Spectrum of the normal EEG signal.

P1: FJT/GAP P2: GYK/GAQ QC: FpQ
Journal of Medical Systems [joms] PP405-368576 March 23, 2002 11:32 Style file version Nov. 19th, 1999

Comparison of Wavelet Transform and FFT Methods in the Analysis of EEG Signals 245

In Fig. 3 it can be observed that there are several peaks which have abnormal
amplitudes in the δ and θ frequency bands. These peaks may indicate a pathological
case such as epilepsy, tumors, and traumas.
In these two figures, the difference of the dominant frequencies can easily be
seen. This means that by using Fourier transform, the frequency components that
the signal includes can be estimated.

Wavelet Transform
Wavelet transform decomposes a signal onto a set of basis functions called
wavelets. These are obtained from a single prototype wavelet, called mother wavelet,
by dilatations and contractions, as well as by shifts.(8)
The first recorded mention of the term “wavelet” was in 1909, in a thesis by
Alfred Haar. The concept of wavelets in its present theoretical form was first pro-
posed by Jean Morlet and the team at the Marseille Theoretical Physics Center,
working under Alex Grossmann in France. The methods of wavelet analysis have
been developed mainly by Y. Meyer and his colleagues, who have ensured the
methods’ dissemination. The main algorithm dates back to the work of Stephane
Mallat in 1988. Since then, research on wavelets has become international. Such
research is particularly active in the United States, where it is spearheaded by
the work of scientists such as Ingrid Daubechies, Ronald Coifman, and Victor
Wickerhauser.(9)
On the basis of the input signal x(t), WT may be continuous WT (CWT) or
discrete WT (DWT). The CWT is expressed as
Z
∗
CWT (a, b) = x(t)9a,b (t)dt (3)

where ∗ denotes the complex conjugate, a ∈ R + represents the scale parameter, and
b ∈ R + represents the translation. The function 9a,b(t) is obtained by scaling the
prototype wavelet 9(t) at time a and scale b, and is defined as
µ ¶
1 t −b
9a,b(t) = √ 9 . (4)
a a
Generally in wavelet applications, orthogonal dyadic functions are chosen as
the mother wavelet. This transform is often discretized in a and b on a dyadic
grid, with the time remaining continuous. The commonly used mother wavelet is
defined as
9 j,k(t) = 2− j/2 9(2− j t − k) (5)
where {9 j,k(t), j, k ∈ Z } is for L2 (R).
Wavelet transforms of the EEG signals in Figs. 1and 2 are shown in Figs. 5 and 6.
In these figures EEG signal is separated into its δ, θ , α, and β bands by using WT.
Observing these figures, it can be easily seen that the amplitudes of signals in the δ and
θ ranges of Fig. 5 is greater than those of Fig. 6. Thus in Fig. 5 the amplitude values of
the δ subspectral component are significantly high. This means that an abnormality
occurs in this section of the EEG signal.
P1: FJT/GAP P2: GYK/GAQ QC: FpQ
Journal of Medical Systems [joms] PP405-368576 March 23, 2002 11:32 Style file version Nov. 19th, 1999

246 Akin

Fig. 5. Wavelet transform of the pathological EEG signal.

Also by using WT, one can view the shapes of the subspectral components of
the EEG signal in time domain to be different from those in Fourier transform.

RESULTS AND CONCLUSIONS

This study is aimed at the comparison of the FFT and WT. Therefore, a better-
suited method for the analysis of EEG signals is investigated.
The fact is that EEG signals are nonstationary, and by the use of the Fourier
transform, small changes may not be realized and the analysis my change depending
on the length of data. So, for spectral analysis, it can be said that WT is more suitable
than Fourier transform. The reason of this success depends on the scaling and the
shifting properties of the mother wavelet.
Another advantage of the wavelet transform is 3D representation of signals
as amplitude, frequency, and time. The 3D representation is more convenient for
pathological cases, as the epileptic discharges using WT subspectral components can
also be represented as shown in Figs. 5 and 6.
In addition to these above expressions we can say that WT is a newer technique,
and also that WT has been developed depending on the short-time Fourier transform.
Namely, WT uses the FFT by adding the time-domain viewment.
 P1: FJT/GAP P2: GYK/GAQ QC: FpQ
Journal of Medical Systems [joms] PP405-368576 March 23, 2002 11:32 Style file version Nov. 19th, 1999

Comparison of Wavelet Transform and FFT Methods in the Analysis of EEG Signals 247

Fig. 6. Wavelet transform of the normal EEG signal.

As a result, one can say that in the spectral analysis, WT is much more efficient
than the FFT.

REFERENCES

1. Akin, M., Investigation of Excited Brain Potential With Spectral Analysis Methods, Ph.D. Thesis,
University of Erciyes, Turkey, 1995.
2. Arserim, M. A., The Estimation of Brain Signals With Modern Spectral Analysis Methods, Master
Thesis, Diyarbakir, Turkey, 2001.
3. Bruce, A., Donoho, D., and Gao, H.-y., Wavelet analysis. IEEE Spectrum 26, 1996.
4. Akin, M., Kiymik, M. K., Application of periodogram and AR spectral analysis to EEG signals.
J. Med. Syst. 24(4), 2000.
5. Cromwell, L., Weibell, F. J., and Pfeiffer, E. A., Biomedical Instrumentation and Measurements,
Prentice-Hall, Englewood Cliffs, NJ, 1980.
6. Proakis, J., and Manolakis, D., Digital Signal Processing, Prentice-Hall, Englewood Cliffs, NJ, 1996.
7. Akin, M., Kiymik, M. K., Arserim, M. A., and Turkoglu, I., Separation of Brain Signals Using FFT
and Neural Networks, Biyomut, Istanbul, Turkey, 2000.
8. Hazarika, N., Classification of EEG signals using the wavelet transform. Signal Processing 59(1):61,
May 1997.
9. Misiti, M., Misiti, Y., Oppenheim, G., and Poggi, J. M., Wavelet Toolbox Users Guide, The MathWorks,
Inc. 1996–1997.
10. Goswami, J., and Chan, A., Fundementals of Wavelets, Wiley, New York, 1999.

View publication stats