Cognitive Brain Research 20 (2004) 256 – 260

www.elsevier.com/locate/cogbrainres

Research report
Alpha power dependent light stimulation: dynamics of event-related
(de)synchronization in human electroencephalogram
Michael Woertz a, Gert Pfurtscheller a,b,*, Wolfgang Klimesch c
a
Department of Medical Informatics, Institute of Biomedical Engineering, University of Technology Graz, Inffeldgasse 16a, Graz 8010, Austria
b
Ludwig Boltzmann Institute for Medical Informatics and Neuroinformatics, Graz 8010, Austria
c
Department of Physiological Psychology, University of Salzburg, Salzburg 5020, Austria
Accepted 18 March 2004

Available online

Abstract

The desynchronization and resynchronization of alpha oscillations was studied in 10 normal subjects after visual stimulation of both eyes
under two experimental conditions, ‘‘eyes opened’’ and ‘‘eyes closed’’. The electroencephalogram (EEG) was recorded bipolarly over the
occipital area and sampled at 200 Hz. The data was processed in real time and evaluated online. In accordance with the alpha power in the 7
to 13 Hz band, short red light flashes of 10 ms duration were presented at intervals of at least 2 s. This stimulation resulted in an event-related
desynchronization (ERD) followed by resynchronization. Trials were controlled for artifacts, averaged offline, and the amount of event-
related (de)synchronization was calculated. The event-related desynchronization was significantly larger in the eyes open paradigm. In
addition, the latencies of event-related desynchronization and resynchronization maxima were larger in the eyes open paradigm compared to
the eyes closed one.
D 2004 Elsevier B.V. All rights reserved.

Theme: Sensory systems

Topic: Subcortical visual pathways

Keywords: Alpha rhythm; Visual stimulation; Event-related synchronization; Event-related desynchronization; Feedback stimulation; EEG

1. Introduction Numerous studies have been performed relating presti-

It is well known that the posterior alpha rhythm is
temporarily blocked or desynchronized by visual stimula-
tion [1,3]. Although it has been demonstrated that desynch-
ronization depends on several factors including reference or
resting power, task difficulty, and type of cognitive
demands, little is known about the dynamics of desynchro-
nization with respect to natural fluctuations in alpha activity
under real-time conditions. In brain computer interface
(BCI) experiments, the EEG is analyzed in real time [10].
In a typical design, subjects have to respond mentally to a
visual cue stimulus. It is therefore important to know
whether the processing of a cue is influenced by ongoing
changes in a subject’s EEG alpha activity.
* Corresponding author. Tel.: +43-316-873-5300; fax: +43-316-873-
5349.
E-mail address: pfurtscheller@tugraz.at (G. Pfurtscheller).
mulus (reference) power to task performance. However,
online stimulation paradigms, in which the ongoing EEG
is scanned throughout the recording session and stimuli are
applied directly depending on determined properties like,
e.g. alpha amplitude, have rarely been used (e.g. Refs. [2,8]).
We investigated the influence of differences in reference
power on the dynamics of alpha power dependent light
stimulation. The influence of large reference power was
analyzed in an orthogonal design. We distinguished between
(i) large versus small reference alpha power using subgroups
of trials of each subject (analyzed offline), and (ii) light
stimulation during eyes open and eyes closed. The latter
condition is also of importance for BCI research, because in
some types of tasks ongoing EEG activity is also analyzed in
periods with closed eyes.
Alpha power was measured continuously and a visual
stimulus was presented only after the power exceeded a
predetermined threshold during a reference interval. The

0926-6410/$ - see front matter D 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cogbrainres.2004.03.014
 M. Woertz et al. / Cognitive Brain Research 20 (2004) 256–260
threshold for stimulation was determined individually be-
fore subjects performed the eyes open and eyes closed alpha
dependent stimulation task.
2. Materials and methods
Ten healthy subjects, six males and four females, vol-
unteered in this study. After the experiment had been
explained to them, subjects gave their informed written
consent. They were seated in a darkened room. Modified
welding goggles with five red light emitting diodes (ar-
ranged in the form of a cross) on each side were used for
stimulation with 10 ms single flashes. Occipital electroen-
cephalogram (EEG) was measured with two gold scalp
electrodes taking the bipolar difference O1
O2 (distance
5 cm) with reference on the forehead (Fz): (O1
O2) versus
Fz. The EEG signal was filtered between 0.5 and 30 Hz and
sampled at 200 Hz with 50 Hz notch filtering enabled. The
real-time alpha oscillation detection was based on band-pass
filtering in the 7 –13 Hz band using a second order Butter-
worth filter approximation. Bandpower values were calcu-
lated using a time window of 250 ms with overlap N-1 (49
samples), which means that for each new filtered EEG
sample a new bandpower value was determined. In order
to trigger a stimulus, the bandpower had to exceed a
predetermined threshold continuously for 250 ms. The
threshold was calculated online from the first stimulation-
free minute of each measurement and then set to the mean
bandpower value plus two times its standard deviation.
Subjects were asked to leave their eyes open during this
reference period. After determination, this threshold was
kept constant for the rest of the run. The principle of this
closed loop EEG processing can be viewed in Fig. 1.
In the experiments, two EEG recordings, each of which
lasted 11 min, were taken from subjects. The very first
minute of each recording served to calculate the stimulation
threshold. Then, alpha power dependent stimulation was
carried out in the remaining 10 min. In the first recording,
Fig. 1. Principle of closed loop data processing and online stimulation. The
EEG signal taken from the subject is first amplified and then fed to a real-
time processing system. The online calculations performed yield the
stimulation signal for the stimulation device (modified welding goggles,
equipped with light emitting diodes).
257
Fig. 2. Principle of ERD/ERS calculation using the intertrial variance
method. First, all trials t(i, j) (i being the trial-index and j the sample-index)
have to be filtered in the selected frequency band. Second, the average over
all trials has to be calculated. The subtraction of this mean trial from every
trial in the third step, squaring, gives the method its name ‘‘intertrial
variance’’. Phase locked events, e.g. evoked potentials, are thereby discarded
from the final result. After averaging the squared trials, ERD/ERS is
determined by using a value Tref, which is normally the mean calculated
from a reference interval before the task, respectively, stimulation started.
subjects were asked to leave their eyes opened during this
10-min period with stimulation. In the second recording,
subjects were instructed to keep their eyes closed during the
10 min with alpha power dependent stimulation. A refrac-
tory period after each stimulation was established to last at
least 2 s. The stimulation signal was recorded and used as a
trigger to divide the filtered EEG data into trials lasting 6 s.
Artifacts were screened visually. On average, the measure-
ment resulted in 84 trials for the eyes open and 93 trials for
the eyes closed experiment. The calculation of subject
specific bandpower and event-related desynchronization/
event-related synchronization (ERD/ERS) results were then
based on the intertrial variance method [4,7]. The principle
of this method is outlined in Fig. 2. All trials were band-pass
filtered in the 7 –13 Hz band. Afterwards, the average of
these filtered trials was calculated and in the following step
subtracted from each trial separately, which omitted phase
locked events like, e.g. visual evoked potentials. After
squaring and averaging over trials, smoothing with a second
order low-pass filter gave the bandpower results. In order to
calculate ERD/ERS percentages, a reference interval (
1.5
to
0.5 s) before stimulation had to be chosen. The mean
of this interval served as reference value to calculate:
T̄f ðjÞ
T̄ref
ERD=ERS%ðjÞ ¼ ð1Þ
T̄ref
where T̄f( j) represents the average over trials with ( j) being
the sample index and T̄ref the determined reference value.
258 M. Woertz et al. / Cognitive Brain Research 20 (2004) 256–260

Fig. 3. (a,b) Results of the online stimulation after averaging over all subjects. Left panel (a) shows bandpower values in the range from 7 to 13 Hz, right panel
(b) indicates the percentages relative to the reference interval from
1.5 to
0.5 s before stimulation. Thick lines indicate the ‘‘eyes closed’’ experiment
contrasting the ‘‘eyes opened’’ measurement. (c) Outlined curves of measured values for both conditions (eyes opened and eyes closed) with respect to
reference power (small and large). Only the square roots of the mean for reference power, for maximum and minimum peaks and for maximal
resynchronization are given as measured with their timings. Other values in between these characteristic points are linearly interpolated.

For a statistical evaluation, we calculated analyses of latency in a time window of 0 – 500 ms poststimulus, (ii)
variances (ANOVAs) for the following EEG measures (cf. minimum power (‘desynchronization’) peak and latency (in
Fig. 3c), determined individually for each subject and each a time window of 0– 2000 ms poststimulus), (iii) maximal
trial, (i) maximum power (‘synchronization’) peak and resynchronization and latency (in a time window of 0– 2000
 M. Woertz et al. / Cognitive Brain Research 20 (2004) 256–260
ms poststimulus) and (iv) maximal ERD for minimum
power. EEG measure (iv) is different from (ii) as the
calculation of ERD scales the change in power. With the
calculation of percentages, a high reference and high event-
related power gives the same results like a low reference and
low event-related power.
3. Results
The average absolute bandpower waveforms of all sub-
jects and the ERD/ERS curves are depicted for the eyes
open and eyes closed condition in Fig. 3a,b.
The data show a large increase in power with a maximum
at stimulus presentation and a sharp decline immediately after
stimulation under both experimental conditions. The sharp
‘synchronization peak’ is the result of alpha power dependent
feedback stimulation, i.e. a stimulus is presented only during
periods of rising alpha power and after a threshold is
exceeded. Although maximal power during stimulation
appears similar in both situations, the extent of desynchroni-
zation and the latency of the resynchronization reveal clear
differences between eyes open and closed. Because these
differences may be simply due to the larger reference alpha
power level with closed eyes, we differentiated between trials
with small and large reference power (using the mean over all
trials in each condition and subject as a cutoff score to
generate groups of trials falling below and above the respec-
tive median). Then, ANOVAs with the within subject’s
factors condition (COND: eyes open versus closed) and
reference power (REF: small versus large) were calculated
for the seven dependent measures described above: power
and latencies of synchronization (i), desynchronization (ii)
and resynchronization (iii) maxima giving six measures, as
well as the maximal ERD for desynchronization (iv). Fig. 3c
indicates the measured values (i) –(iii). Given is the square
root of the mean from trials with high and low reference
power averaged over all subjects in both conditions, eyes
opened and closed. In Table 1, standard errors of the mean of
measured peaks and latencies are summarized.
The results indicate that neither the synchronization peak
nor its latency showed any significant effects of COND or
Table 1
Standard error of mean (S.E.M.) values for peaks and latencies
Reference Synchronization Sync
(AV) peak (AV) latency (s)
Small reference, 4.885 14.468 0.007
eyes open
Large reference, 7.853 14.654 0.005
eyes open
Small reference, 5.421 13.107 0.008
eyes closed
Large reference, 8.835 13.401 0.007
eyes closed
Given are the S.E.M. values for reference interval, synchronization, desynchronization and resynchronization maxima as depicted in Fig. 3c. S.E.M. values of
peak latencies are given as well.
259
REF. Significant effects for factor COND were obtained for
the desynchronization peak ( F(1,9) = 14.47; p < 0.01) and its
latency ( F(1,9) = 7.15; p < 0.05) indicating that the peak has
smaller power and occurs later in the eyes open as compared
to the eyes closed condition. Factor REF also reached
significance for the desynchronization peak ( F(1,9) = 7.85;
p < 0.05), but not for its latency, showing that trials with
small reference power have a peak with significantly smaller
power than trials with large reference power. A significant
influence from factor COND could also be observed for
maximal resynchronization ( F(1,9) = 9.23; p < 0.05) and its
latency ( F(1,9) = 5.37; p < 0.05) revealing that power during
resynchronization is larger and occurs earlier in the eyes
closed than in the open condition. No significant interaction
between COND and REF could be observed in any of the
above cases. Finally, for maximal ERD, significant F-values
were observed for all three variance sources, COND
( F(1,9) = 10.60; p < 0.01), REF ( F(1,9) = 25.49; p < 0.01),
and COND  REF ( F(1,9) = 9.83; p < 0.05). These findings
indicate that desynchronization is larger during eyes open
and depends on reference power in that large reference
power leads to the strongest event-related desynchronization
in the eyes open condition.
4. Discussion
For the closed loop approach (i.e. the alpha power
dependent stimulation) used in the present study, we found
different reactivity patterns for light stimulation during
closed and opened eyes. Whereas desynchronization latency
varies only as a matter of stimulation condition (eyes open
versus closed) and is generally reduced during eyes closed,
power estimates vary as a function of stimulation condition
and reference power as well. For ERD, the findings reveal
an interesting dissociation with respect to reference power
and stimulation condition. First, for the eyes open condition
(with small overall reference power) ERD is significantly
larger than for the eyes closed condition (with large overall
reference power). Second, the influence of large versus
small reference power has a significantly stronger enhancing
influence on ERD for the eyes open condition. In other
Desynchronization Desync Resynchronization Resync
peak (AV) latency (s) maximum (AV) latency (s)
2.203 0.023 8.175 0.161
2.734 0.032 8.154 0.176
4.466 0.032 9.159 0.084
5.265 0.031 11.564 0.112
260 M. Woertz et al. / Cognitive Brain Research 20 (2004) 256–260
words, there is a general tendency that large reference power
is related to a larger extent of desynchronization (ERD). In
studies focusing on cognitive performance, it was shown
that large reference power and a large ERD are closely
associated with good performance, whereas small reference
power and a small ERD are related to comparatively bad
performance (for a review see Ref. [5]). Although we are not
able to make inferences about cognitive performance in the
present study, the finding that ERD depends on the presti-
mulus reference power is in good line with the general
phenomenon that already the prestimulus reference has an
impact on the extent of ERD. The potential meaning of the
present finding for cognitive processes may be interpreted
on the basis of the significant interaction between COND
and REF with ERD as a dependent measure. If the extent of
desynchronization in the present paradigm would be influ-
enced only by the larger stimulus intensity during eyes
open, ERD should be not differentially affected by reference
power. Thus, not only stimulus intensity, but also the
cognitive state, too (eyes closed may be related to a more
drowsy state than eyes open), has an impact on the extent of
ERD. The fact that alpha reactivity (as measured by ERD) is
larger for eyes open than closed may also be interpreted in
terms of the stronger stimulus intensity in the former
condition. The differential enhancing effect of reference
power, however, provides evidence that this latter interpre-
tation may not generally be valid. We assume that the
particularly strong enhancing effect of reference power on
ERD during eyes open (as evidenced by the significant
interaction between COND and REF) reflects the influence
of a different (more attentive) cognitive state, rather than the
influence in stimulus intensity.
The timing of visual cue presentation with respect to
alpha amplitudes is of interest in BCI studies. After present-
ing the cue indicating that the system is ready to accept
mentally induced EEG changes from a user, online classi-
fication should be made as fast as possible. Here, the results
indicate that a visual stimulus presented to the subject’s eyes
during periods of high alpha activity causes an ERD, which
returns to the baseline level after seconds. This desynchro-
nization is not useful and carries no information for BCI
control [6]. Therefore, it would be necessary to take care not
to present a cue at the same time with high alpha amplitudes
or to delay a possible classification by the alpha resynchro-
nization latency, especially if information from alpha band
is one of the features used to control the BCI.
Finally, it should be noted that our findings may also be
explained in part by the differential influences of the parvo-
and magnocellulary system. It is most likely that light
stimulation with open eyes primarily activates the parvocel-
lular system whereas during closed eyes, the magnocellular
system will be primarily activated. The reason for this
assumption is that with open eyes, the perception of lines,
contours, and colors is possible which is analyzed by the
parvocellular system whereas with closed eyes, changes in
brightness are the only feature that can be analyzed. The
important point here is that the magnocellular system
responds earlier and faster than the parvocellular system
(for a review cf. Ref. [9]). In agreement with this notion, the
findings demonstrate that the latencies of the de- and
resynchronization maxima are significantly reduced during
stimulation with closed eyes. Future research, however, is
necessary to further validate this interpretation.
Acknowledgements
This work was supported in part by the ‘‘Fonds zur
Förderung der wissenschaftlichen Forschung’’, Project
P14831.
References
[1] E. Adrian, B. Matthews, The Berger rhythm: potential changes from
the occipital lobes in man, Brain 57 (1934) 355 – 385.
[2] E. Basar, E. Rahn, T. Demiralp, M. Schürmann, Spontaneous EEG
theta activity controls frontal evoked potential amplitudes, Electro-
encephalogr. Clin. Neurophysiol. 108 (1998) 101 – 109.
[3] H. Berger, Über das Electroenkephalogramm des Menchen, Arch.
Psychiatr. Nervenkr. 87 (1929) 527 – 570.
[4] J. Kalcher, G. Pfurtscheller, Discrimination between phase-locked and
non-phase-locked event-related EEG activity, Electroencephalogr.
Clin. Neurophysiol. 94 (1995) 381 – 384.
[5] W. Klimesch, EEG alpha and theta oscillations reflect cognitive and
memory performance: a review and analysis, Brain Res. Rev. 29
(1999) 169 – 195.
[6] C. Neuper, G. Müller, A. Kübler, N. Birbaumer, G. Pfurtscheller,
Clinical application of an EEG-based brain-computer interface: a case
study in a patient with severe motor impairment, Clin. Neurophysiol.
114 (2003) 399 – 409.
[7] G. Pfurtscheller, Quantification of ERD and ERS in the time domain,
in: G. Pfurtscheller, F.H. Lopes da Silva (Eds.), Event-related
desynchronization, Revised edition, Handbook Electroencephalogr.
Clin. Neurophysiol. vol. 6, Elsevier, Amsterdam, 1999, pp. 89 – 105.
[8] E. Rahn, E. Basar, Enhancement of visual evoked potentials by stim-
ulation during low prestimulus EEG stages, Int. J. Neurosci. 72
(1993) 123 – 136.
[9] R. Silberstein, Steady-state visually evoked potentials, brain resonan-
ces and cognitive processes, in: P. Nunez (Ed.), Neocortical dynam-
ics and human EEG rhythms, Oxford Univ. Press, New York, 1995,
pp. 272 – 303.
[10] J. Wolpaw, N. Birbaumer, D. McFarland, G. Pfurtscheller, T. Vaughan,
Brain-computer interfaces for communication and control, Clin. Neu-
rophysiol. 113 (2002) 767 – 791.