ELSEVIER Electroencephalography and clinical Neurophysiology 94 (1995) 381-384

o
Short c o m m u n i c a t i o n

Discrimination between phase-locked and non-phase-locked

event-related EEG activity
J. Kalcher, G. Pfurtscheller *
Department of Medical lnformatics, Institute of Biomedical Engineering, Graz University of Technology, and Ludwig Boltzmann Institute
for Medical Informatics and Neuroinformatics, Brockrnanngasse 41, A-8010 Graz, Austria
Accepted for publication: 9 February 1995

Abstract

Differentiation b e t w e e n phase-locked and non-phase-locked event-related E E G activity is an important task in the evaluation o f
event-related E E G activity. Event-related changes o f E E G activity such as event-related desynchronization (ERD) or event-related
synchronization (ERS) can be quantified by either instantaneous band p o w e r or intertrial variance calculations. In the former case the
E R D or ERS can be masked by an event-related potential while in the latter it is not. E x a m p l e s from sensory stimulation and m o v e m e n t
experiments, where the E R D (ERS) is calculated by both methods, are s h o w n and discussed.

Keywords: Event-related desynchronization (ERD); Event-related synchronization (ERS); Intertrial variance; Event-related potential

1. Introduction ERS. Phase-locked (e.g. evoked potentials) and non-phase-locked EEG

activities cannot be discriminated when both types of activity are within
Phase-locked EEG activities include all types of event-related poten- the same frequency band. This means an ERP can mask EEG desynchro-
tials (ERPs) such as evoked potentials and slow negative potential shifts. nization or synchronization. One method to overcome this problem is to
For the quantification of these types of activity the additive averaging calculate band power changes by using the intertrial variance of bandpass
technique can be used, whereby the non-phase-locked EEG activity is filtered EEG data. The aim of this study is to demonstrate the advantage
attenuated. Non-phase-locked EEG activities also include event-locked of using both methods of ERD (ERS) calculation based on event-related
(stimulus-locked) oscillations such as induced alpha or beta band activi- EEG data.
ties. Typical examples are the oscillations around 40 Hz after visual
stimulation (Eckhorn et al., 1988; Gray and Singer, 1989), the induced
mu rhythm over the hand area during visual stimulation (Koshino and
2. Methods
Niedermeyer, 1975; Pfurtscheller and Neuper, 1994) and the phasic
oscillations around 40 Hz before different types of movements
(Pfurtscheller et al., 1993). Event-related EEG trials (e.g. stimulus synchronous or movement
The term event-related desynchronization (ERD) is used to describe synchronous) can be processed in different ways (Fig. 1):
the event-related, short-lasting and localized amplitude attenuation of (i) Additive averaging of N trials results in the ERP and the intertrial
EEG rhythms within the alpha or beta band, while event-related synchro- variance (ERP calculation).
nization (ERS) describes the event-related, short-lasting and localized (ii) Bandpass filtering of each trial, squaring of samples and averaging
enhancement of these rhythms (Pfurtscheller and Aranibar, 1979; of N trials results in a time course of instantaneous band power (classical
Pfurtscheller, 1992). The events giving rise to such frequency band ERD method, Power method (P)).
specific changes of amplitudes in the spontaneous EEG can be either (iii) Bandpass filtering of each trial and calculating the intertrial
externally (any kind of stimulation) or internally (e.g. voluntary move- variance by averaging of all trials also results in a time course of
ment) paced. With the "classical" ERD method (Pfurtscheller and instantaneous band power (alternative method of ERD processing, intro-
Aranibar, 1979; Boiten et al., 1991; Klimesch et al., 1992; Derambure et duced in this paper, intertrial variance (IV) method).
al., 1993), event-related EEG data are first bandpass filtered, the samples Using additive averaging (i) of a number of trials would lead to a
squared and then averaged over trials. A decrease of band power is an reduction of non-phase-locked components in the signal and therefore
indicator for an ERD, and an increase of band power an indicator for an make it possible to discriminate between event-related and non-event-re-
lated changes of bioelectrical activity. The calculation of the ERP by
additive averaging is as follows:
li=N
* Corresponding author. Tel.: AUT (+316) 82 16 94; Fax: AUT ERP(j) = ~ .'~'-'~x(i, j)
(+316) 81 29 64. ~1

0013-4694/95/$09.50 © 1995 Elsevier Science Ireland Ltd. All rights reserved

SSDI 0013 -4 694(95 ) 0 0 0 4 0 - 2 EEG 94653
382 J. Kalcher, G. Pfurtscheller / Electroencephalography and clinical Neurophysiology 94 (1995) 381-384

where N = total number of trials, and x(i,j) = j-th sample of the i-th trial tribute to the band power changes, while in the latter method only
of the data. non-phase-locked activity is quantified.
For the classical ERD method (ii), the instantaneous power is calcu- The significance of a band power decrease (increase) at each sample
lated as: point (or average of sample points) of the time course can be tested by
1 t~
applying the following sign test:
~(J) = ~ , E Xf2(i,j) k01 ~ x
l=1
where P(j) = averaged power estimation of bandpass filtered data (aver- i=l~ -

aged over all trials), xf(i,j)=j-th sample of the i-th trial of the bandpass
where N = total number of trials and ktj ) = number of trials having a
filtered data, while for the intertrial variance method (iii), the variance is
power decrease (increase) at the j-th sample.
calculated as:
The sign is evaluated from the change of power (either decrease or
1 i=N increase) at the j-th sample point in each single trial, compared to the
IV(j) = ~ - ~ - 1__~
1"= {xf(i,j) _ ~f(j)}2 median of the power sample in the reference interval of that trial. By
using the median of the reference period instead of the mean, the impact
where N = total number of trials, xf(i,j)=j-th sample of the i-th trial of of outliers is reduced. The results of this test can be displayed in the form
the bandpass filtered data and ~f(i)= mean of data at the j-th sample of a step function and give an indication of the significance of the ERD
(averaged over all bandpass filtered trials). (ERS).
For the ERD calculation, either the instantaneous power (ii) or the
intertrial variance (iii) can be used. The ERD is quantified as the
percentage change of the power or intertrial variance (A(j)) at each 3. R e s u l t s
sample point or an average of some samples (in this paper samples are
averaged in time windows of 125 msec) relative to the average power or To demonstrate the differences between ERD calculation using either
intertrial variance (R) in a reference interval: method (ii) or (iii), 3 examples are given. In the first experiment the
R - A(j) subject was stimulated by single flashes every 5 sec and EEG was
ERD(J) ~ × 100% recorded from the positions Cz-A1 according to the international 10/20
system and sampled at 128 Hz. The flash stimuli resulted in both a visual
1 n°+k
evoked potential (VEP) and a blocking response of the alpha rhythm (10
with R = ~ . E A0) Hz ERD); the former is a phase-locked phenomenon and the latter is not.
J=n 0
In Fig. 2a the VEP with the intertrial variance is plotted. The graphs in
where R = average power in reference interval, averaged over k samples, Fig. 2b-g show band power time courses of the 71-trial experiment with
and A(j) = power or intertrial variance at the j-th sample. a sample rate of 4/sec, with the two upper graphs displaying data in the
The difference between methods (ii) and (iii) is that in the former frequency range of 6-12 Hz, the middle graphs in the frequency range of
method both phase-locked and non-phase-locked EEG activities con- 8-12 Hz and the lower graphs in the frequency range of 4 - 7 Hz. The left

I event-related EEGtrials ]
x(i,j)
~ ba-~ndpass
~ filter
Xf(i,j) }
squaring
xf=(i,j) 1 I squaring
(Xf(i,j) -~--f(j))'

I I
averaging ( }
overtrials i averaging
overtrials I averaging
overtrials
I I

powermethod intertrlalvariance
method
Fig. 1. Methods for processing event-related EEG trials. Averaging of trials of raw EEG data (data vector x) results in an ERP and averaging of bandpass
filtered and squared EEG data (data vector y z) or calculation of intertrial variance of bandpass filtered EEG data (data vector y - ~) results in an ERD
(ERS) time course.
 J. Kalcher, G. Pfurtscheller / Electroencephalography and clinical Neurophysiology 94 (1995) 381-384 383

diagrams were calculated by method (ii) and the right by method (iii). made discrete finger flexions at intervals of about 10 sec. In both
The power peaks in the left diagrams represent a contribution from the examples the EEG was recorded over the left hand area (electrode C3,
EP power; these peaks are not present in the right diagrams. As can be nose reference) and was sampled at 128 Hz. ERS time courses were
seen from Fig. 2 the VEP contributes dominantly in the 4 - 7 Hz band and calculated in the 36-40 Hz band using both methods (ii) and (iii) with a
to a smaller extent in the 6 - 1 2 Hz band when the ERD is calculated in sample rate of 4/sec.
the classical way using method (ii). No difference is evident between the For the first experiment, a marked difference was found between the
two methods in the 8 - 1 2 Hz band. results for the two different calculation methods, with a reduced power
Fig. 3 displays data from 2 other experiments. In the first, a subject peak found for the intertfial variance method. For the movement experi-
was stimulated electrically at the right wrist with a rectangular pulse of ment, the results were identical for both the power and intertfial variance
60 mA and a duration of 0.1 msec. In the second experiment, a subject methods. This can be interpreted to mean that electrical stimulation

-25-

I a
i
O-

2 5

~ h ~ ~ ~
• s e c
stimulation .~
~" ~?_~H3 ~ b 25 ~ H , z
0
C
--

--'75
:i ~.,~. _

~
I_~
-25

-- ~~
~0.

"72 ~
--10C -- n, - - ~ , , 2
o ~ ~ ~ ~ 0 1 S~C 3 4

5 0 ~ e
:~5- ~O.o Hz d =~ ~2.o Hz
0
c! ~ , 0 e
- : ~ 5

- 5 0

-75 _~.5 -2
-1 O01
l
072 ~ ~1 ~ 4
_ 1 o o . ~ ' 2
0 ~
._
~
. .
~ ~
_~

5o~ 50 ¸ V

25

0
7_~0

=v
!

...
25-

0
H z, g
-25- 23.5 -25. 22.~J~
-50- -50"

-75- -75"
72 72
-100 --2 --100
0 -i 2 ~ 4 0 -i :~ ~ 4
~ e c s e c

Fig. 2. Examples of processed EEG trials from a VEP experiment: VEP and intertrial variance (a), ERD time courses calculated by the power method in
different frequency bands (b, d, f) and corresponding ERD time courses calculated by the intertrial variance method (c, e, g). Note the differences between
phase-locked and non-phase-locked data; only in the case of the power method (left side) does the ERP contribute to the ERD. The reference interval is
indicated by 2 short vertical lines (0.5-1.5 sec). The scale on the left marks the percentage power increase and decrease and the scale on the fight the
significance levels (from 10-2 to 10-s). On the left side of each diagram the bandpass frequency, average power in the reference interval and number of
trials (from top to bottom) are indicated.
384 J. Kalcher, G. Pfurtscheller / Electroencephalography and clinical Neurophysiology 94 (1995) 381-384

=oo.p 2 0 0 V

150

1 OI:~" 100
a
~0-
36.0
4 0 . 0
..~,0 o :oOHz/ b
,~^ , ~.~,~/~_~/-~' -8 0 ~,, • ^....,/it ~ , , / " ' / 8
0 ~- ~ v - ~v - I - ~,..¢~.~ ~ ' V -- |

-~o ~. ~ -50 1.1 I -.4

9"1 I -2
_~ OOo9 ~ ~ ~ ~ -1 O0
0 -i 6, :5 ~. ~ 6
5 6
& sec ~j~ :~ e¢~:

stimulation stimulation

c d
Hz :,~0

8 (~
~
_F
--50 O. 7 ]~ - 5 0 ~.,7 l~- _
1 ~ o E J --: "1 O 0
~2o rq ~ e ~
-w oo -i
o ~ ~ 3 ,n. ~ 1~ 7 8 i ~ ~ 4 ~ 6 ~ 8
~, m;~(:: ~ ~e(::

movement movement
Fig. 3. ERS time course from the stimulation experiment (a, b) and the movement experiment (c, d). Data on the left are calculated by the power method
and on the right by the intertrial variance method. Note that in the first case phase-locked signal components contribute to the ERS, which are eliminated
using the IV method. In the second case no ERPs are found, which leads to similar ERS curves in both methods.

results in a stimulus-locked artefact in the 40 Hz range, whereas in a Derambure, P., Defebvre, L., Dujardin, K., Bourriez, J.L., Jacquesson,
movement task, only induced 40 Hz oscillations are present with a J.M., Destee, A. and Guieu, J.D. Effect of aging on the spatio-tem-
dominance prior to movement onset. poral pattern of event-related desynchronization during voluntary
movement. Electroenceph. clin. Neurophysiol., 1993, 89: 197-203.
Eckhorn, R., Bauer, R., Jordan, W., Brosch, M., Kruse, W., Munk, M.
4. Discussion
and Reitboeck, H.J. Coherent oscillations: a mechanism of feature
The computation of the ERD (ERS) according to the two different linking in the visual cortex? Biol. Cybern., 1988, 60: 121-130.
methods shows that a discrimination between phase-locked and non- Gray, C.M. and Singer, W. Stimulus-specific neuronal oscillations in
phase-locked EEG activities can be easily made. When lower frequency orientation columns of cat visual cortex. Proc. Nat. Acad. Sci. (USA),
bands are analyzed the ERD (ERS), calculated by the power method, can 1989, 86: 1698-1702.
be influenced by ERP components which have their mean energy in the Kaufman, L., Schwartz, B., Salustri, C. and Williamson, S.J. Modulation
lower alpha and theta bands. When the upper alpha band is analyzed this of spontaneous brain activity during mental imagery. J. Cogn. Neu-
influence can be ignored. Kaufman et al. (1989) used the calculation of rosci., 1989, 2: 124-132.
the intertrial variance of bandpass filtered MEG data to quantify "alpha Klimesch, W., Pfurtscheller, G. and Schimke, H. Pre- and post-stimulus
field suppression" which is a synonym for the ERD within the alpha processes in category judgement tasks as measured by event-related
band. The calculation of band power changes by both methods is of desynchronization (ERD). J. Psychophysiol., 1992, 6: 185-204.
special importance when activity in the upper beta (gamma) band is Koshino, Y. and Niedermeyer, E. Enhancement of rolandic mu-rhythm by
investigated and allows the discrimination of stimulus- and non-stimulus- pattern vision. Electroenceph. clin. Neurophysiol., 1975, 38: 535-538.
locked components. The examples in Fig. 3c,d display induced, non- Pfurtscheller, G. Event-related synchronization (ERS): an electrophysio-
phase-locked oscillations around 40 Hz, typical for the preparation of logical correlate of cortical areas at rest. Electroenceph. clin. Neuro-
movement (Pfurtscheller et al., 1993). physiol., 1992, 83: 62-69.
Pfurtscheller, G. and Aranibar, A. Evaluation of event-related desynchro-
nisation (ERD) preceding and following voluntatry self-paced move-
Acknowledgements
ment. Electroenceph. clin. Neurophysiol., 1979, 46: 138-146.
We would like to thank Roland Sagmeister for programming. The Pfurtscheller, G. and Neuper, C. Event-related synchronization of mu
research was supported by the °'Fonds zur F6rderung der wis- rhythm in the EEG over the cortical hand area in man. Neurosci.
senschafllichen Forschung" Project P10000 and the "Oesterreichische Lett., 1994, 174: 93-96.
Nationalbank" Project 4534, Pfurtscheller, G., Neuper, C. and Kalcher, J. 40-Hz oscillations during
motor behavior in man. Neurosci. Lett., 1993, 164: 179-182.
Toro, C., Cox, C., Friehs, G., Ojakangas, C., Maxwell, R., Gates, J.R.,
References
Gumnit, R.J. and Ebner, T.J. 8-12 Hz rhythmic oscillations in human
Boiten, F., Sergeant, J. and Geuze, R. Event-related desynchronization: motor cortex during two-dimensional ann movements: evidence for
the effects of energetic and computational demands. Electroenceph. representation of kinematic parameters. Electroenceph. clin. Neuro-
clin. Neurophysiol., 1991, 82: 302-309. physiol., 1994, 93: 390-403.