Anda di halaman 1dari 47

EEG and VEP signal processing

Daniel Novák, Lenka Lhotská, Vladimír Eck, Milan Sorf,

Department of Cybernetics, Czech Technical University in Prague, Czech Republic

Daniel Novák, Department of Cybernetics, Faculty of Electrical Engineering, Czech

Technical University in Prague, Karlovo namesti 13, Prague 2, 121 35, Email:

Abstract: The basic algorithms for Electroencephalography signals (EEG) as well as

for Visual Evoked Potentials (VEP) are presented. The methods have been applied
using the EEG data set of more than twenty students during practical course Cognitive
Processes. Each students has processed its recorded file using basic algorithms such as
Fourier, wavelet transform, spectral analysis, mapping and spectral coherence. The goal
of the work is to pre-process EEG data for feature extraction and consequently
performing classification human states in the frame of Intelligent Human Computer
Interface (iHCI). The report can also serve as a teaching material where apart from the
descriptions of different algorithms the physiological background of EEG and EP is also
depicted. In the last part the overall achieved results are summarized and the
conclusions are drawn with the main focus on iHCI.
Keywords: Electroencephalography signals, signal processing, evoked potentials,
human machine interface

CONTENT....................................................................................................................... 1
1. INTRODUCTION ...................................................................................................... 2
2. BRAIN FUNDAMENTALS....................................................................................... 3
2.1 HOW THE BRAIN WORKS?......................................................................................... 3
2.2 BRAIN GEOGRAPHY ................................................................................................. 5
3. WHAT IS EEG? ......................................................................................................... 8
3.1 MEASURING THE EEG ............................................................................................. 8
2.1 ELECTROENCEPHALOGRAPHY WAVEFORMS............................................................. 9
2.1.1 Frequency ...................................................................................................... 10
2.1.2 Morphology.................................................................................................... 11
4. EEG SIGNAL PROCESSING................................................................................. 16
4.1 FOURIER TRANSFORM ............................................................................................ 16
4.2 WAVELET TRANSFORM .......................................................................................... 19
4.3 EEG BRAIN TOPOGRAPHY...................................................................................... 20
4.3.1 Spectral Coherence........................................................................................ 25
4.4 INDEPENDENT COMPONENT ANALYSIS (ICA)........................................................ 26
4.5 OTHERS ALGORITHMS FOR EEG PROCESSING ........................................................ 28
5. EVOKED POTENTIALS ........................................................................................ 29
5.1 WHAT ARE ERPS? ................................................................................................. 29
5.2 HOW ARE ERP S USED ? .......................................................................................... 29
5.3 VISUAL EVOKED POTENTIALS ............................................................................... 29
5.3.1 Factors influencing VEP................................................................................ 30
5.3.2 Measurement set-up....................................................................................... 31
5.3.3 VEP signal processing ................................................................................... 32
5.4 BRAINSTEM AUDITORY- EVOKED POTENTIAL......................................................... 37
6. CONCLUSIONS ....................................................................................................... 39
7. REFERENCES ......................................................................................................... 40
APPENDIX I ................................................................................................................. 43
APPENDIX II................................................................................................................ 45

EEG and VEP signal processing 1

2. Brain Fundamentals

1. Introduction
The EEG is still one of the most main tools to access to one of the most
unknown and complex system is nature. There are no doubts that due to its complexity
and ability to reflect underlying processes in the brain the EEG signal is theoretically
the best physiological signal for extraction and comprehension of human behaviour.
The majority obtained results serve as basic pre-processing for Intelligent
Human-Computer Interface (iHCI) that mainly focuses to personality type classification
problem [Sorf 2001], [Lhotska 2000]. In that context the pre-processing and feature
extraction play crucial role in a satisfactory performance of the proposed classification
system [Janku 2000], [Sorf 2000]. These and related issues are currently under
investigation [Eck 2000].
The following report summarize the main used techniques in EEG signal
processing field. Apart from EEG analysis the processing of Visual Evoked Potentials
(VEP) is treated here as well. The gathered examples are based on the laboratory
experiments realized during the subject Cognitive Processes. The experiment was
divided into two parts. In the first part the EEG signal of young male persons was
recorded with the emphasis on artificial artefact generation such as blink, muscle and
teeth squeeze, etc. Considering VEP measurement the set of more than twenty student
was stimulated by the pattern reversal of the full-field chessboard in the biological
laboratory. Each eye was tested separately. Afterward all data were processed using
standard techniques described below.
At the same time this material can be used as a tutorial introduction of the
covered problematic for the course Cognitive Processes. Therefore we take into
consideration the multiply purpose of the work resulting in clear and schematic
description rather than detail topics exploration and analysis.
The rest of report is organized as follows. Section 2 looks at the structure of the
brain system and its anatomy. In section 3, the EEG characteristics will be described.
Section 4 provides algorithms survey of EEG processing. Section 5 describes in general
manner evoked potentials, followed by section 6 that explains into more details theory
of Visual Evoked Potentials. Finally section 7 includes discussions and concluding
remarks. Since the EEG processing is one of the elements in the iHCI mosaics we
consider important to briefly present philosophy behind human interactions systems in
appendix II.

EEG and VEP signal processing 2

2. Brain Fundamentals

2. Brain Fundamentals
EEG (electroencephalogram) reflects electrical activity of a multitude of neural
populations in the brain. This signal is extremely complex, since EEG is generated as a
superposition of different simultaneously acting dynamical systems. Before we proceed
to description of different algorithms used in signal-processing society we will briefly
explain some basic medicine aspects of the brain itself. We will start from the most
simple “dynamical system” –neuron.

2.1 How the brain works?

A basic insight into the physical mechanics of the brain may assist in
understanding how potential differences are developed giving rises to various EEG
techniques [Fisk_WWW].
The human brain contains approximately 100 billion neurons or
nerve cells. From the cell body containing the nucleus, branches
called ‘processes’ trail in all directions. The nerve cells longest
process (its ‘main cable’) is the axon, which carries outgoing signals
(see Figure 1). An axon may extend all the way from the Central
Nervous System CNS (which consists of the brain and spinal cord)
to a finger or toe, connecting with the muscle on which it acts.

The Resting Membrane Potential

Neurons send messages through an electrochemical process. When

chemicals in the body are electronically charged, they are called
ions. The important ions in the CNS are sodium and potassium (both
having a single positive charge), calcium (having two positive
charges) and chloride (having a single negative charge). There are
also some negatively charged protein molecules. Nerve cells are
surrounded by a membrane that allows some ions to pass through
whilst blocking the passage of other ions. This type of membrane is
termed semi permeable. When a neuron is not sending a signal it is
said to be ‘at rest’. When a neuron is at rest, the inside of the neuron
is negative relative to the outside. The concentrations of the
different ions attempt to balance out on both sides of the membrane,
but never reach an even state of density due to the cell membrane
allowing only specific ions to pass through channels (ion channels).
At rest, potassium ions (K+) can cross through the membrane easily
whereas chloride ions (Cl-) and sodium ions (Na+) are resisted. The
negatively charged protein molecules (A-) inside the neuron are also
resisted by the cell membrane. In addition to these selective ion
channels, there is a ‘biological pump’ that uses energy to transfer
three sodium ions out of the neuron for every two potassium ions it
Figure 1.
receives. Finally, when all these forces are at equilibrium, the
A neuron with
trailing axon potential difference between the inside and outside of the neuron is
approximately -70 mV. This is the resting membrane potential of a

EEG and VEP signal processing 3

2. Brain Fundamentals

The Action Potential

The action potential indicates what happens when a neuron transmits information
from one cell to another. The action potential is an explosion of electrical activity that is
created by a depolarising current. This means that a stimulus causes the resting potential
to move toward 0 mV (see Figure 2). When the depolarisation reaches about -55 mV, a
neuron’s threshold, it will fire an action potential. If the neuron does not reach this
critical threshold level, no action potential will fire. Furthermore, when the threshold
level is reached, an action potential of fixed magnitude will always fire. So for any
given neuron, the size of the action potential is always the same. This is called the ‘all
or none’ principle.

Figure 2. An action potential

A simple analogy between digital logic gates and neurons can be drawn:
Neurons can act as switches or logical decision units directing the flow of information.
Depending upon the pattern of signals arriving at its ‘Synapses’ (nerve cell’s input
junctions) , a neuron either does or does not send new signals along its axon. Thus, in
principle, the brain can be thought of as a network of interconnected decision-making
The ‘cause’ of the action potential is an exchange of ions across the neuron
membrane. A stimulus first results in the opening of sodium channels. Since there are a
larger number of sodium ions on the outside of the cell and that the inside is negative
relative to the outside, sodium ions rush into the neuron via the membrane. This results
in the neuron becoming more positive and consequently it depolarises. There is a small
latency before the potassium channels open, allowing K+ ions to leave the cell,
reversing the depolarisation. It is at this time that the sodium channels start to close.
This causes the action potential to return to -70 mV (a repolarisation). The action
potential in reality exceed -70 mV (a hyperpolarisation) since the potassium channels
remain open for longer. Gradually, the ion concentration returns to equilibrium and the
cell to -70 mV (see Figure 3).

EEG and VEP signal processing 4

2. Brain Fundamentals

Figure 3. The chemical attributes of an Action Potential

2.2 Brain Geography

A working knowledge of various brain functions and their associated locations
may prove valuable in determining electrode positions for EEG recordings.
The brain is encased in bone and further protected by three encircling
membranes, ‘the meninges’. The Cerebrospinal fluid (CSF) which circulates the
cerebral ventricles helps to cushion delicate nerve tissue; it is these forms of protection
to the brain that the electrical signals must pass through before reaching an electrode.

The brain is made up of four main structures:

• Cerebrum
• Cerebellum
• Pons
• Medulla Oblongata
The Cerebrum is the largest area of the brain. It consists of two hemispheres,
the right and left cerebral hemispheres. The right cerebral hemisphere controls the left
side of the body and the left cerebral hemisphere the right. The outer layer of the
cerebrum, called the cerebral cortex, is made up of grey matter. The inner portion of the
cerebrum is white matter. Grey matter is composed of nerve cells. These cells control
brain activity. White matter is composed of nerve cell axons that carry information
between nerve cells in the brain and spinal cord.

Deep indentations called fissures divide each hemisphere of the cerebrum into four
lobes – see Figure 4:

• Frontal Lobe
• Parietal Lobe
• Temporal Lobe
• Occipital Lobe

EEG and VEP signal processing 5

2. Brain Fundamentals

Figure 4. Brain basis four lobes

The Cerebellum is the second largest area of the brain. It consists of two
hemispheres connected by the vermis. The cerebellum connects to the brain stem. The
Cerebellum, together with the thalamus and cerebrum, controls skilled muscular co-

Figure 5. Brain locations of the Cerebellum, Pons, Medulla and Lobes.

The Pons (Figure 5) co-ordinates the activities of the cerebrum and cerebellum by
relaying impulses between them and the spinal cord. The pons contains the origins of
the 5th , 6th , 7th and 8th cranial nerves (see Table 1).

EEG and VEP signal processing 6

2. Brain Fundamentals

Table 1. There are 12 pairs of cranial nerves:

1 Olfactory Smell
2 Optic Vision
3 Oculomotor Eye movement
4 Trochlear Eye movement
5 Trigeminal Facial sensation
6 Abducent Eye movement
7 Facial Face movement
8 Vestibulocochlear Hearing and balance
9 Glossopharyngeal Taste
10 Vagus Involuntary muscles
Voluntary neck
11 Accessory
12 Hypoglossal Tongue movement

The Medulla Oblongata controls respiration and heat beat. It connects the brain
with the spinal cord. It contains the origins of the 9th , 10th , 11th and 12th cranial nerves.
Besides these four main structures (Cerebrum, Cerebellum, Pons, Medulla
Oblongata), many other specialised nerve structures help make up the brain (Figure 6):
• Hypothalamus - Controls water balance, sleep, temperature, appetite and blood
• Thalamus - Monitors input from the senses and acts as a relay station for the
sensory centre of the cerebrum.
• Limbic System - Together with the hypothalamus controls hunger, thirst,
emotional reactions and biological rhythms.
• Brain Stem - Controls basic functions, including blood pressure, heart beat and

Figure 6. Brain locations of various nerve structures

EEG and VEP signal processing 7

3. What is EEG?

3. What is EEG?
EEG is spontaneous cortical electrical activity recorded at the scalp. The human
EEG was discorvered by Berger in the 1800s using a primitive galvanometer with a
surface electrode placed on his son’s scalp and recorded a rhythmic pattern of electrical
oscillation. This signal was the moment-by-moment electrophysiological response of
cortical brain cells. It is now thought that the electrical potentials recorded as EEG are
produced by electrical dipoles in the pyramidal cell layer. Many Pyramidal cells and
their dendrites are arranged vertically. This arrangment sets up a dendro-somatic
“dipole” or potential which oscillates with the arrival of excitatory or inhibitory
postsynaptic potentials [Bates_WWW].

3.1 Measuring the EEG

The electrical activity of the brain behaves like any other material system.
Changes in membrane polarisation, inhibitory and excitatory post synaptic potentials,
action potentials etc. impress voltages are conducted throught the surrounding medium
of CSF, meninges, skull and scalp.
The electrical voltages conduct up through brain tissue, enter the membranes
surrounding the brain, continue on up through the skull to appear at the scalp. At this
point they are reduced from the milliVolt range (of the membrane potential gradients
and action potentials) to a few microVolts. That is, the voltage is measured in millionths
of a Volt.
Typical values might be 20-100 microVolts for EEG with lesser values being
recorded in averaged evoked potentials: perhaps 10 microVolts. Larger values are
recorded in epilepsy and other disorders which can be diagnosed using EEG.
These potentials are easily recorded. An electrode of some conductive metal
(gold, silver-silverchloride, and tin are popular choices) is attached to the scalp in a bed
of conductive gel (salt solution). Various methods exist to accomplish this. The
electrodes (at least two are required to measure EEG) are fed into a sensitive amplifier.
The EEG signal is small compared to the amplitude of common artifacts
(muscle, mains power frequency radiation). Clean signals are dependent on low
scalp/electrode impedance, differential amplifiers, and filtering.
The EEG is typically recorded by an “instrumentation” amplifier which uses a
third “common” electrode to remove noise. The signals on this common channel is
subtracted from each of the signal and reference channels. This removes signals which
are present on both the signal and reference electrodes (the common mode signal).
These are then compared. A good amplifier has a high comon mode rejection ratio - the
ability to subtract out these common signals. This is measured in dB and should be
more 90db.
Additional factors in amplifier design include high input impedance. This
determines the amount of current that an impressed voltage will drive. High input
impedance leads to a low current beign drawn and, consequently, even a high resistance
scalp-electrode interface will be quite immune to picking up the environmental
electrical noise which surrounds us. An electrically shielded recording laboratory can
also help in this respect.

EEG and VEP signal processing 8

3. What is EEG?

EEG is most often recorded from many electrodes in a arranged in a particular

pattern or montage. A common standard for describing these position is the
International 10/20 System that has been used as well in our case-see Appendix I.
These methods are cheap and give a continuous record of brain activity with
better than millisecond resolution. No other tool can achieve this high temporal
resolution and for this reasons many of the detailed discoveries of dynamic cognitive
processes have been reported using EEG and ERP (Event Related Potentials-see section
5.1) methods.
In our experiments the electroencephalogram was recorded by means of
Beckmann Ag/AgCl-electrodes attached to the scalp of the subject with Beckmann EEG
electrode paste. The skin was prepared beforehand by rubbing with a cleaning paste.
After a few minutes the electrode resistance was measured and was usually below 2 kW.

2.1 Electroencephalography waveforms

Electroencephalography waveforms generally are classified according to their
frequency, amplitude, and shape, as well as the sites on the scalp at which they are
recorded. The most familiar classification uses EEG waveform frequency (eg, alpha,
beta, theta).
Information about waveform frequency and shape is combined with the age of
the patient, state of alertness or sleep, and head site to determine significance.
In the following Figure 7 is depicted EEG of a young healthy person.

Figure 7. EEG example of a young person

Normal EEG waveforms are defined and described by the following criteria
according to [Louis_WWW 01]:
• Frequency (Hertz, Hz) is the initial characteristic used to define normal or abnormal
EEG rhythms.
• Most waves of 7.5 Hz and higher frequencies are normal findings in the EEG of an
awake adult. Waves with a frequency of 7 Hz or less often are classed as abnormal
in awake adults, although they normally can be seen in children or in adults who are
asleep. In certain situations, EEG waveforms of an appropriate frequency for age
and state of alertness are considered abnormal because they occur at an
EEG and VEP signal processing 9
3. What is EEG?

inappropriate scalp location or demonstrate irregularities in rhythmicity or

• Some waves are recognized by their shape, head distribution, and symmetry. Certain
patterns are normal at specific ages or states of alertness and sleep.
• The morphology of a wave may resemble specific shapes, such as vertex (V) waves
seen over the vertex of the scalp in stage 2 sleep or triphasic waves that occur in the
setting of various encephalopathies.
As a result we can divide EEG signal into two groups: accrding to its frequency
context and morphology charescteristics.

2.1.1 Frequency
Most waves range from 0.5-500 Hz, but most clinical EEGs have been
performed on paper-writing machines with upper ranges of 20-40 Hz.
• Alpha waves - 8-13 Hz
• Beta waves - Greater than 13 Hz
• Theta waves - 3.5-7.5 Hz
• Delta waves - 3 Hz or less

Alpha waves

Alpha waves generally are seen in all age groups but are most common in adults.
They occur rhythmically on both sides of the head but are often slightly higher in
amplitude on the nondominant side, especially in right-handed individuals. They tend to
be present posteriorly more than anteriorly and are especially prominent with closed
eyes and with relaxation. Alpha activity disappears normally with attention (eg, mental
arithmetic, stress, opening eyes). In most instances, it is regarded as a normal
waveform. An abnormal exception is alpha coma, most often caused by hypoxic-
ischemic encephalopathy of destructive processes in the pons (eg, intracerebral
hemorrhage). In alpha coma, alpha waves are distributed uniformly both anteriorly and
posteriorly in these patients, who are unresponsive to stimuli.

Figure 8. Alpha wave

Beta waves

Beta activity is ‘fast’ activity. It has a frequency of 14 and greater Hz. It is

usually seen on both sides in symmetrical distribution and is most evident frontally. It is
accentuated by sedative-hypnotic drugs especially the benzodiazepines and the
barbiturates. It may be absent or reduced in areas of cortical damage. It is generally
regarded as a normal rhythm. It is the dominant rhythm in patients who are alert or

EEG and VEP signal processing 10

3. What is EEG?

anxious or who have their eyes open. Beta waves are observed in all age groups. They
tend to be small in amplitude and usually are symmetric and more evident anteriorly.

Figure 9. Beta wave

Theta waves

Theta activity has a frequency of 3.5 to 7.5 Hz and is classed as “slow” activity.
It is abnormal in awake adults but is perfectly normal in children upto 13 years and in

Figure 10. Theta wave

Delta waves

Delta activity is 3 Hz or below. It tends to be the highest in amplitude and the

slowest waves. It is quite normal and is the dominant rhythm in infants up to one year
and in stages 3 and 4 of sleep. Delta waves are abnormal in the awake adult. Delta
waves can be focal (local pathology) or diffuse (generalized dysfunction). Theta and
delta waves are known collectively as slow waves.

Figure 11. Theta wave

2.1.2 Morphology
This section identifies some normal waveforms, including K complex, V waves,
lambda waves, positive occipital sharp transients of sleep (POSTS), spindles, mu

EEG and VEP signal processing 11

3. What is EEG?

rhythm, spikes, sharp waves, and certain delta waves (polyphasic and monophasic
These waves are recognized by their shape and form and secondarily by their
frequency. They include waves that may be normal in some settings and abnormal in
others (eg, spikes, sharp waves). The next discussion of EEG morpohology follows
Spike and wave format is seen at all ages but most often in children. It consists
of a spike, which is probable generated in the cortex, and a large amplitude slow wave
(usually delta), thought to originate from thalamic structures, occuring recurrently. They
may occur synchronously and symmetrically in the generalized epilepsies or focally in
the partial ones. In the generalized types of spike and wave, true absense (petit mal) is
characterized by 3 Hz spike-wave, while slow spike-wave occurs more usually with
brain injury and the Lennox-Gastaut syndrome.


Polyspike and wave is a form of spike wave in which each slow wave is
accompanied by two or more spikes. The usual pattern is that the spike and wave is
faster than 3 Hz - usually 3.5 to 4.5 Hz. It is often associated with myoclonus or
myoclonic seizures. It should not be confused with 6 Hz spike and wave, otherwise
known as phantom spike and wave - a normal variant.


Lambda and POSTS are similar morphologically, and have a triangular
shape.They occur posteriorly and symmetrically. POSTS stands for ‘positive occipital
transients of sleep’ and occurs in stage 2 sleep. Lambda occurs in the awake patient
when the eyes stare at blank surfaces. Both are normal wave forms and can occur singly
or in long or short runs.

K Complexes occur in sleep when arroused - thus K complexes are seen with
noises or other stimuli especially in stage 2 sleep. The K complex is often followed by

EEG and VEP signal processing 12

3. What is EEG?

an arrousal response - namely a run of theta waves of high amplitude. Following this the
EEG shows sleep again or the awake state.

V waves occur in the parasaggital areas of the two sides and take the form of
sharp waves or even spikes which show in the biparietal regions (vertex) withphase
reversal at the midline in tranverse montages or at the vertex in front-to-back ones. They
are seen in stage 2 sleep along with spindles, K complexes, POSTS, etc..

Mu activity is a rhythm in which the waves have a shape suggestive of a wicket
fence with sharp tips and rounded bases. It may show phase reversal between two
channels. The frequency is generally half of the fast activity present.

Psychomotor variant is a rare rhythm which appears to be an harmonic of two or
more basic rhythms causing a complex form. As can be seen it is higher in amplitude
than the surround and the waves have a notched appearance. It is quite assymetrical and
is often mistaken for paroxysmal activity. It is benign. It is also known as Fourteen and
Six Rhyrhm.


Fourteen and six activity is most often seen in children and adolescents. As seen
it takes the form of 6 Hz and 14 Hz waves sometimes going in the same direction (up
or down) and in others in opposite directions. It is typically seen in sleep or drowsiness
and is usually seen in monopolar recordings.

EEG and VEP signal processing 13

3. What is EEG?


PLEDS - Periodic Lateralized Epileptiform Discharges - are a form of discharge
associated with acute brain injury or damage. The pattern is also said to be most evident
when acute brain injury is coupled with some metabolic derangement. It is an evolving
pattern starting out with sharp waves occuring at regular intervals over one whole area
or side with relative flattening between. It evolves to slow transients and then slow
waves occuring periodically with improving background activity between. It finally
clears completely. It is often associated with severe focal signs and much illness early
on with some improvement later.

Triphasic waves are as illustrated the 3 waves are seen outlined in white. They
often occur as long runs causing an appearance of pseudoparoxysmal activity. The
waveform was originally found with hepatic encephalopathy but has subsequently been
found in association with many other forms of metabolic encephalopathy.

Burst-suppression is a pattern of burst of slow and mixed waves often of high
amplitude alternating with a flat baseline. The pattern is bilateral but not always
symmetrical. It is usually seen after severe brain injury such as postischemia or
postanoxia. It is also seen in temporary form in deep anesthesia in a stage prior to total
flattening of the EEG.

EEG and VEP signal processing 14

3. What is EEG?

Artifacts are waves or groups of waves which are produced by technical or other
disturbances which are not due to brain activity. The massive amplification magnifies
all manner of disturbances such as EKG and Pulse artifacts, electrode and movement,
and IV and 50 Hz artifacts and Sweat artifacts.
Sweat artifactwhich represent salt solution between electrodes shorting them out.


Both artifacts are illustrated and are recognized by their periodicity. The EKG
shows periodic QRS complexes as the EKG is a much larger electrical signal than EEG.
Pulse artifact is caused by a pulse under an electrode moving that electrode periodically.
Both are easily recognized usually but can be a problem.


Patient movement artifacts are abrupt with a rapid upstroke in almost all
instances. They are large in amplitude and last a long time by EEG standards. A ‘POP’
is a brief electrode shift which can be mistaken for a spike by the naive but it is seen
usually in only two adjascent channels and not three as. Occurs with epileptic spikes.


These artefacts are seen commonly in Intensive Care Unit recordings and both
are electrical interference. The ‘drip’ artifact is shown as the first; it is periodic and
small in amplitude and easily recognized. Sixty Hz is seen where the electrode contacts
are poor, grounding is inadequate and electrical contrivances are in use close by. This
causes spikes at 50 per second.

EEG and VEP signal processing 15

4. EEG signal processing

4. EEG signal processing

The complex character of EEG and its significance in brain research and clinical
practice brought early introduction of signal analysis methods to EEG studies. In the
next subsections we will briefly describe basic methods for EEG signal processing. We
will start with the spectral methods. Spectral methods of EEG processing can be traced
back to the first attempt of Fourier analysis application to EEG in 1932 [Dietsch 1932].

4.1 Fourier Transform

In 19th century, the French mathematician J. Fourier, showed that any periodic
function can be expressed as an infinite sum of periodic complex exponential functions.
Many years after he had discovered this remarkable property of functions, his ideas
were generalised to first non-periodic functions, and then periodic or non-periodic
discrete time signals. It is after this generalisation that it became a very suitable tool for
computer calculations. In 1965, a new algorithm called Fast Fourier Transform (FFT)
was developed and FT became even more popular [Polikar_WWW]. One example of
FT analysis can be seen in Figure 12
The definition of FT is given by:

∫ f (t)e
− jwt
F ( w) = dt
∞ (1)
f (t ) = ∫ F (w)e dw


The information provided by the integral, corresponds to all time instances, since
the integration is from minus infinity to plus infinity over time. This is why Fourier
transform is not suitable if the signal has time varying frequency, i.e., the signal is non-
stationary. This means that the FT tells whether a certain frequency component exists
or not. This information is independent of where in time this component appears.

Figure 12. Fourier transform of all 19 signal appearing in figure Figure 20 . Note how
visible is the 50Hz noise in the spectrum

EEG and VEP signal processing 16

4. EEG signal processing

Therefore a linear time frequency representation called Short Fourier Transform

(STFT, sometimes called Gabor analysis) was introduced. In STFT, the signal is divided
into small enough segments, where these segments (portions) of the signal can be
assumed to be stationary. For this purpose, a window function is chosen. The width of
this window must be equal to the segment of the signal where its stationarity is valid.
The following definition of the STFT summarises all the above explanations in one line:
STFT ( l , w) = ∫ [ f (t ) w* (t − l )]e − jwt dt (2)

where w is a window function.

Limitations of FFT and STFT promoted introduction of parametric methods

such as autoregressive (AR) or autoregressive - moving average (ARMA) models, free
from the windowing effects. Most of the used methods in EEG processing, linear and
non-linear, assume stationarity of the signal-in spite of the fact, that information
processing by brain is mostly reflected in fast dynamic changes of its activity. This fact
implies the application to the analysis of EEG and LFP (local field potentials) methods
operating in time-frequency space.


It is an extension of FT transformed into 3D space. The spectrum in each time-interval

instance is sorted along time axis to provide compression visual outluk of EEG signal
behaviour. In the figure Figure 13 CAS is computed for four electrodes. The rectangle
snapshots an blink artefact that is evident in time space (Figure 14).

Figure 13. Example of CSA for detection of blink artefact

EEG and VEP signal processing 17

4. EEG signal processing

Figure 14. The blink artefact appearance in time space

Detail examination of Figure 13 that is demonstrated in Figure 15 reveals the repeated
pattern-appearance of blink artefact.

Figure 15. The detection of blink artefact. On each 15 Hz frequency is a periodic peak
in spectrum corresponding to the artefact.

EEG and VEP signal processing 18

4. EEG signal processing

4.2 Wavelet Transform

There are two main differences between the STFT and the CWT:
1. The Fourier transforms of the windowed signals are not taken
2. The width of the window is changed as the transform is computed for every
single spectral component, which is probably the most significant characteristic
of the wavelet transform.
The continuous wavelet transform (CWT) is defined as follows [Daubechies 1992]:
C (a , b) = ∫ f (t )ψ a ,b (t ) dt (3)


ψ a, b (t ) = a 2ψ ( )
a (4)
is a window function called the mother wavelet a is a scale and b is a translation.
Using discrete wavelet transform (DWT) we can avoid time consuming
computation of wavelet continuous transform. The DWT is defined is follows[Burros
1991] where f(n) is discrete function:
C (a , b) = C ( j , k ) = ∑ f (n )ψ j, k ( n) (5)
n∈ Z

where ψ j ,k is a discrete wavelet defined as:

ψ j, k ( n) = 2 − j / 2ψ ( 2 − j n − k ) (6)
The parameters a,b are defined in such a way that a = 2 j , b = 2 j k . Sometimes
the analysis is called dyadic as well. The inverse transform is defined in a similar way
like [Burros 1991]:
f (n ) = ∑∑ C( j , k )ψ j , k ( n) (7)
j ∈ Z k ∈Z

To be useful, wavelet theory must come with fast algorithms for machine
computation, that is, a method like FFT both for finding the wavelet coefficients
C ( j , k ) and for reconstructing the function they represent. There is an even faster
family of algorithms based on a completely different idea, namely that of
multiresolution analysis. We suggest the work [Mallat 1989] to an interested reader.
One example of applying DWT is depicted in Figure 16.

EEG and VEP signal processing 19

4. EEG signal processing

Figure 16. Typical example of EEG discrete wavelet transform including electrode
arefact (wavelet function sym5, decomposition level 6).
Linear decomposition of signal in a wavelet basis was a significant improvement
over the short time Fourier transform, allowing for orthogonal representation, fast
numerical implementations and multiresolution decomposition of signals [Mallat 1989].
WT is being successfully applied e.g. for analysis of time-locked EEG phenomena
(evoked potentials), where it’s main drawback-sensitivity of the representation to the
time shift of analyzed window [Durka 1996] is not essential. However, neither WT nor
STFT provide enough resolution and flexibility in a general case, like description of
transients occurring more or less randomly in time.

4.3 EEG brain topography

With the possibility of recording simultaneously a great number of digitized
channels of EEG, a new technique was born: EEG brain topography, at the end of the
80s. In this technique a large number of electrodes is placed onto the head, following a
geometrical array of even-spaced points. A special software inside the apparatus’
computer, plots the activity on a color screen or printer, by coding the amount of
activity in several tones of color (for example, black and blue might depict low EEG

EEG and VEP signal processing 20

4. EEG signal processing

amplitude, while yellow and red might depict larger amplitudes). The spatial points
lying between electrodes are calculated by mathematical techniques of interpolation
(calculating intermediary values on the basis on the value of its neighbors), and thus a
smooth gradation of colors is achieved [Renato_WWW].

Figure 17. The detection of teeth squeeze artefact using amplitude mapping

Figure 18. The corresponding amplitude map

This approach gives a much more accurate and representative view of the
location of alterations of rhythm, amplitude, etc., in relation to i the surface of the skull.
Neurologists working with the EEG brain topographic system were soon able to
differentiate several kinds of diagnoses (including some mental diseases whose
biological origin, or ethiology, was previously unknown). Pinpointing the exact location

EEG and VEP signal processing 21

4. EEG signal processing

of EEG alterations was also made much more easier. In addition, the use of the cinè
mode (animations using several sequential pictures taken from the brain maps) made
possible the dynamic study of brain function in action [Maurer 1991].

Figure 19. Schematic illustration of brain mapping interpolation using enhanced 10-20
system-see Appendix I
EEG brain topography is not performed in all cases requiring a recording of the
brain activity. Its main indication is to determine the presence of tumors and focal
disease of the brain (including epilepsy, arteriovenous mal-formations and stroke). It is
also appropriate when disturbances in consciousness and vigilance are present, such as
narcolepsy (the abrupt onset of sleep), coma, etc.
In addition, EEG brain topography is being increasingly used to monitor the
effects of withdrawal of psychoactive drugs, and in infectious diseases of the braim,
such as meningites, as well as to follow up patients who where subjected to brain
operations. In psychiatry, EEG brain topography has been of value in identifying
disorders of biological origin, such as schizophrenia, dementias, hyperactivity and
depression, brain atrophy and attention deficit disorders in children [Peter 1995].

EEG and VEP signal processing 22

4. EEG signal processing

It depends where the interpolation is performed. If we take directly amplitude values in

time space we obtain amplitude map. This approach is revealed by Figure 18 in which
the teeth squeeze artifact from Figure 17 is analyzed. The schematic illustration how the
amplitude interpolation works is demonstrated in Figure 19.

Figure 20. The electrode artefact in time space

Considering the interpolation implied in frequency space we get so called
spectral map. We return again to electrode artefact detection. Consequent three Figures
21,22,23 uncovers the situation before, during and after artefact (Figure 20).

EEG and VEP signal processing 23

4. EEG signal processing

Figure 21. Spectral map before electrode artefact

Figure 22. Spectral map during electrode artefact

Figure 23. Spectral map after electrode artefact

EEG and VEP signal processing 24

4. EEG signal processing

4.3.1 Spectral Coherence

A special results of spectral analysis is the measure of coherence between two
electrodes. It assesses the similarity of spectral content of two electrodes over time and
is usually taken to reflect a measure of “coupling” between brain regions. It is virtually
impossible to estimate coherence by visual EEG inspection. Some illnesses may begin
with abnormalities of cortical coupling. Leuchter has reported such abnormalities in
Alzheimer’s disease and Thatcher found abnormality of coherence as the best
discriminator of mild closed head injury [Hardvard_WWW].
We will show the abilities of this techniques on the example of blink artefact
detection (Figure 24). We will implement the spectral coherence calculation in the four
frequency bands corresponding to the definition of alpha, beta, delta and game waves-
see section 2.1.2 MorphologyThe final result is shown in Figure 25.

Figure 24. EEG signal in time, the section before (tags 1,2) during (tag 3,4) and after
blink artefact (tags 5,6) are displayed.

Figure 25. Spectral Coherence maps after, during and before blink artefacts.

EEG and VEP signal processing 25

4. EEG signal processing

4.4 Independent Component Analysis (ICA)

Independent Component Analysis [C OMMON 1994] was originally proposed to
solve the blind source separation problem, to recover N source signals, s={s1 (t), …,
sN(t)}, (e.g., different voice, music, or noise sources) after they are linearly mixed by
multiplying by A, an unknown matrix, x = {x 1(t), …, x N(t)} = As, while assuming as little
as possible about the natures of A or the component signals. Specifically, one tries to
recover a version, u=Wx, of the original sources, s, identical save for scaling and
permutation, by finding a square matrix, W, specifying spatial filters that linearly invert
the mixing process. The key assumption used in ICA to solve this problem is that the
time courses of activation of the sources (or in other cases the spatial weights) are as
statistically independent as possible.
Mathematically, the ICA problem is as follows: We are given a collection of N-
dimensional random vectors, x (sound pressure levels at N microphones, N-pixel
patches of a larger image, outputs of N scalp electrodes recording brain potentials, or
nearly any other kind of multi-dimensional signal). Typically there are diffuse and
complex patterns of correlation between the elements of the vectors. ICA, like Principal
Component Analysis (PCA), is a method to remove those correlations by multiplying
the data by a matrix as follows:
u = Wx (8)

One example of decomposition (decorrelation) is illustrated in Figure 27.

But while PCA only uses second-order statistics (the data covariance matrix),
ICA uses statistics of all orders and pursues a more ambitious objective. While PCA
simply decorrelates the outputs (using an orthogonal matrix W), ICA attempts to make
the outputs statistically independent, while placing no constraints on the matrix W.

Figure 26. The difference between PCA and ICA on a non-orthogonal mixture of two
distributions that are independent and highly sparse (peaked with long tails). An
example of a sparse distribution is the Laplacian: p(x)=ke|-x|. PCA, looking for
orthogonal axes ranked in terms of maximum variance completely misses the structure
of the data. Although these distributions may look strange, they are quite common in
natural data.
Most ICA is performed using information-theoretic unsupervised learning
algorithms. Despite its relatively short history, ICA is rapidly becoming a standard
technique in multivariate analysis [Jung 2001] .
ICA analysis may be used to segregate obvious artifactual EEG components
(line and muscle noise, eye movements) from other sources. Furthemore ICA analysis is

EEG and VEP signal processing 26

4. EEG signal processing

capable of isolating overlapping alpha and theta wave. ICA appears to be a promising
new analysis tool for human EEG and ERP research. It can isolate a wide range of
artefacts to a few output channels while removing them from remaining channels. These
may in turn represent the time course of activity in long lasting or transient independent
brain sources on which the algorithm con verges reliably. [Scott 1996] has shown that
by incorporating higher order statistical information ICA avoids the non uniqueness
associated with decorrelating decompositions.

Figure 27. Left: 4.5 seconds of EEG data. Right: an ICA transform of the same data
using weights trained on 6 minutes of similar data from the same session.

EEG and VEP signal processing 27

4. EEG signal processing

4.5 Others algorithms for EEG processing

Development of non-linear methods of signal analysis brought explosion of
works concerning application of chaotic formalism to EEG [Basar 1989] [Dvorák
1991]. Works of Freeman, based on carefully planned physiological experiments,
brought new insights into the mechanisms of non-linear modes of brain operation
[Freeman 1987]. However, tracing of chaos in EEG proved to be more difficult than it
was expected. It has been shown, that linear forecasting of signals from human and
animal brains is as good-and sometimes better-as the non-linear prediction [Blinowska
1991]. Tests of non-linear procedures, performed on surrogate (phase disturbed) data,
revealed their failure in description of different kinds of EEG time series [Achermann
et al. 1994], [Pijn et al. 1991].
In view of the presently available evidence, application of typical methods of
non-linear time series analysis, such as calculation of attractor dimension or Ljapunov
coefficients, seems to be fully justified only for special cases such as e.g. epileptic EEG
[Pijn et al., 1991] or activity of a group of neurons in a well defined operation mode
[Skarda and Freeman, 1987]. This fact it is not surprising when we consider the
multitude of dynamic processes running in brain and their ever-changing character.
High resolution time-frequency representation of signal’s energy can be
constructed by Cohen’s class transforms [Williams, 1997]; they all derive from the
quadratic Wigner transform which can be expressed as:

τ τ
W f (t , ω ) = ∫ f ( t + ) f (t − )e −iϖτ dτ (9)
2 2

This representation satisfies the time and frequency marginals, but contains
severe cross terms between different time-frequency structures, which may lead to
misinterpretation. Sophisticated mathematics applied to reduce this effect created class
of Reduced Interference Distributions (see e.g. [Williams, 1997]), where reduction of
cross terms is usually achieved at the cost of marginal properties.
In spite of high resolution offered by Cohen’s class transforms, their application
is practically limited to visual inspection of time-frequency plots for each analyzed data
epoch. The method, which allows for parametric-fully quantitative-description of
signals in time-frequency space is the Matching Pursuit (MP), a method based on
adaptive approximation of time series by functions chosen for each analyzed epoch. The
MP was introduced by [Mallat 1993]. The first application to biological signal
concerned EEG analysis [Durka 1995].

EEG and VEP signal processing 28

5. Evoked Potentials

5. Evoked Potentials

5.1 What are ERPs?

ERPs (event-related potentials or evoked potentials in the case of external
stimulation) are small voltage fluctuations resulting from evoked neural activity. These
electrical changes are extracted from scalp recordings by computer averaging epochs
(recording periods) of EEG time-locked to repeated occurrences of sensory, cognitive,
or motor events. The spontaneous background EEG fluctuations, which are random
relative to when the stimuli occurred, are averaged out, leaving the event-related brain
potentials. These electrical signals reflect only that activity which is consistently
associated with the stimulus processing in a time-locked way. The ERP thus reflects,
with high temporal resolution, the patterns of neuronal activity evoked by a stimulus
[Lesilie 2001].

5.2 How are ERPs used?

Due to their high temporal resolution, ERPs provide unique and important
timing information about brain processing. Mental operations, such as those involved in
perception, selective attention, language processing, and memory, proceed over time
ranges in the order of tens of milliseconds. Most other functional imaging techniques
require the integrating evoked brain activity over many seconds and are thus unable to
capture the time course (or sequence) of these operations. ERP recordings, however,
provide a millisecond-by-millisecond reflection of evoked brain activity. For this
reason, ERPs are an ideal methodology for studying the timing aspects of both normal
and abnormal cognitive processes. On the other hand, ERP data provide less accurate
spatial information than positron emission tomography (PET) or functional magnetic
resonance imaging (fMRI), which lack fine temporal resolution. As a result, ERPs
represent the natural complement of PET and fMRI to study human cognition. Whereas
PET and fMRI can localize regions of activation during a given mental task, ERPs can
help in defining the time course of these activations.
Almost any sensory modality can be tested in theory; however, in clinical
practice there are only a few that are used on a routine basis. The ones most often
encountered are the visual-evoked responses (VEP, both flash and checkerboard types),
short latency somatosensory-evoked responses (SLSEP), and short latency brainstem-
evoked responses BAER [Lesilie 2001].

5.3 Visual Evoked Potentials

The visual-evoked potential (VEP) tests the function of the visual pathway from
the retina to the occipital cortex.The usual waveform that is depicted in figure FFF is
the initial negative peak (N1 or N75), followed by a large positive peak (P1 or P100),
followed by another negative peak (N2 or N145). Maximum value for P100 is 115 msec
in patients younger than 60 years; it rises to 120 msec thereafter in females and the
values are 5 msec longer in males. Even though published norms are available in the
medical literature, it is preferable to have norms for each individual laboratory to
control for lab to lab variability in technique. The norms that we have used appears in
Table 2.

EEG and VEP signal processing 29

5. Evoked Potentials

N75 N135

N75 N135


Figure 28. Example of response to visual stimuli

Considering the morphology the W wave is most often an individual variation,
however decreasing the stimulation frequency from the ubiquitous 2Hz to 1 Hz usually
will convert the W shape into a conventional P100 peak. Check size and alternation rate
is a factor and by changing these parameters one can obtain a W or a conventional P100
response. Large checks tend to produce VEP similar to flash stimulation.
Table 2. Normative values for determination of significant waveforms
Absolute latency Value (ms) Average(ms) Rigth-to left difference (ms)
P 100 98.1 112.1 5.9
N 75 68.8 82
N 135 132.9 70.4
Peak to peak Value (ms) Average (ms)
N75-N135 36.6 65.2
Amplitudes (µV) Value (µV) Right-to left difference (µV)
P100 7.8 1.6
N75/P100 3.5-20.2 0-5.2

5.3.1 Factors influencing VEP

The usual VEP are evoked generally by checkerboard stimulation and because
cells of the visual cortex are maximally sensitive to movement at the edges, a pattern-
shift method is used with a frequency of 1-2 Hz. The size of the checks affects the
amplitude of the waveform and the latency of the P100. In addition, pupillary size,
gender, and age all affect the VEP. Visual acuity deterioration up to 20/200 does not
alter the response significantly; large checks may be required. In some studies, women
have slightly shorter P100 latencies. Sedation and anesthesia abolish VEP. One has to
consider that some subjects by “fixating” beyond the plane of stimulation may alter or
suppress P100 altogether.

EEG and VEP signal processing 30

5. Evoked Potentials

5.3.2 Measurement set-up

Checkerboard pattern (or less often flash) is used as stimulation in the most
cases-see Figure 29. Responses are collected over Oz, O1, and O2 and with hemifield
studies at T5 and T6 electrodes using the standard EEG electrode placement. Monocular
stimulation is used to avoid masking of a unilateral conduction abnormality. Sedation
should not be used and note should be taken of medications that the patient is taking
Testing circumstances should be standardized including seating distance of 70-
100 cm from the monitor screen, giving a check size of approximately 30 seconds of
visual angle. The vision should be corrected as best as possible in case of a visual
problem. Pupil size and any abnormality should be noted. The P100 waveform is at its
maximum in the midoccipital area. Stimulus rates of 1-2 Hz are recommended, filter
setting should be 1-200 Hz bandwidth (outside limit is 0.2-300 Hz).In our case see
Figure 33 where for the convenience of power frequency removal the bandwidth is set
up to 0-50 Hz.
The recording time window (sweep length) is recommended to be 250 msec; 50-
200 responses are to be averaged. A minimum of 2 trials should be given. The responses
are averaged and the P100 positive polarity waveform that appears in the posterior head
region is analyzed. The mean latency is about 100 msec. Normative data should be
assembled on a lab by lab basis. In Table 3 the step by step description of the performed
experiments appears.

Figure 29. One example of visualization pattern, chessboard 8x5

Appropriate placement of recording electrodes on either side of the active
membranes, using differential recording can enhance the signal while rejecting common
mode noise. Possible signal distortions using differential recording necessitate
interpretation of differentially recorded waveforms with care. A given differentially
recorded waveform may result from a variety of waveform combinations to the
differential montage. The most effective means of enhancing signal to noise ratio is
based on the random nature of noise and on the evoked activity being time locked to the
evoking event (e.g., stimulus onset).

EEG and VEP signal processing 31

5. Evoked Potentials

Table 3. The measurement description (each eye was tested separately).

Session Session type Session description Session duration (ms)
1 click session boxes 82550
2 watch session chessboard 32x20, 1Hz 122873
3 watch session chessboard 16x10, 1Hz 122873
4 watch session chessboard 8x5, 1Hz 122873
5 watch session circles 38781
6 watch session chessboard 32x20, 2Hz 95398
7 watch session chessboard 8x5, 2Hz 95398
8 watch session Disks variable
9 watch session chessboard 32x20, 10 Hz 79312
10 watch session chessboard 8x5, 10 Hz 79312
11 click session Rectangles 103674

The setup for recording evoked potentials therefore includes a stimulus

generator, coupled to the averager which it triggers, and electrodes attached to the
subject and connected to differential filters and amplifiers feeding the analog to digital
converter of the averager. The output of the system is a waveform describing the time-
locked voltage changes recorded from the electrodes in response to the stimulus. The
typical example of signal averaging is displayed in Figure 30. For comparison purposes
all cannels of 10-20 system were processed.

5.3.3 VEP signal processing

The electrodes record hundreds, sometimes thousands, of electrical responses
from the patient’s visual nervous system. Therofore it is main importance to average the
data. Signal averaging across periods following a stimulus will not change the signal
embedded in each period, but will diminish the random noise.
VEP activity includes a variety of neural activities, muscle activity (EMG), heart
activity (EKG) and eye-movement artifacts (EOG) from the subject, in addition to
environmental electric noise such as the mains power supply, instrumental noise such as
video monitor raster, stimulus artifacts etc. Thus, surface recorded evoked potentials are
embedded in considerable noise, resulting in a low signal to noise ratio.
Enhancing the signal to noise ratio is performed by filtering the activity from the
electrode to the frequency band of the signal of interest, rejecting frequencies
dominated by noise-this can be seen clearly on the following two consecutive Figure 31
and 32. In the first picture the presence of power 50 Hz frequency is apparent. In the
second picture the noise is removed using elliptic filter shown in Figure 33. The signals
are already averaged.

EEG and VEP signal processing 32

5. Evoked Potentials

Figure 30. VEP Averaging; all channels are averaged for comparison purposes.

Figure 31. Signal with 50hz noise

EEG and VEP signal processing 33

5. Evoked Potentials

Figure 32. Signal with the noise removed

Figure 33. Low pass FIR filter for enhacment signal-to noise ratio. The filters paramets
are following passband: 40Hz, stopband: 50Hz, sampling frequency: 250Hz
Below, in the Figure 34, is a depiction of obtained signal after above mentioned pre-
processing steps.

EEG and VEP signal processing 34

5. Evoked Potentials

Figure 34. Typical result of VEP analysis. The signal is recorded from O1 electrode,
averegized and filtred.
In this phase all method described in section 4 can be basically used for the VEP
processing as well.
We select one example for all of VEP denoising using wavelets techniques.
Wavelet denoising method or nonlinear wavelet filtering is based on taking wavelet
transform of a signal, passing transform through a threshold, which removes the
coefficients below a certain value, then taking the inverse DWT, as illustrated in Figure
35. The method is able to remove noise and achieve high compression ratios because of
the “concetrating” ability of the wavelet transform.

y Decomposition w Thresholding wδ Reconstruction yδ

W Dδ W-1

Figure 35. Basic denoising concept

Using the denoising scheme we analyze the signal shown in Figure 31 where
the apparent presence of power 50 Hz noise can be seen. The results of denoising and
wavelet decomposition is displayed in Figure 36.

EEG and VEP signal processing 35

5. Evoked Potentials

Figure 36. Signal noise removal -electrode O1 using wavelet denoising. Wavelet
function dmey is implemented as a mother wavelet, decomposition level is 5, the soft
thresholding is used in denoising scheme. In the uppermost graph the original signal is
yellowed and the denoised one is red.
In the [Quiroga 2000] VEP wavelet denoising is performed as well, in the
similar manner as in our case. We include Figure 37 where the comparision beetwen
original and denoised signal in continues time-frequency plane is showed. Note the
development of three typical waves N75, P100, N135 in time.

EEG and VEP signal processing 36

5. Evoked Potentials

Figure 37. Contour plot of VEP signal using continuous wavelet transform. After
denoising (right plot) it is possible to see the evolution of the evoked potentials. Here
the symbol N200 corresponds to N75 and P300 to N135.

5.4 Brainstem auditory-evoked potential

Brainstem auditory-evoked potential (BAEP) or brainstem auditory-evoked
response (BAER) measures the function of the auditory nerve and auditory pathways in
the brain stem. The short latency BAER si generally used for clinical purposes. The test
can be performed under sedation or under general anesthesia. Standard broadband click
stimulation is used on the ear tested, while the contralateral ear receives masking noise
of 30-40 dB lesser intensity. Monoaural stimulation is used. The click intensity should
be 65-70 dB above click perception threshold. A repetition rate of about 10 Hz should
be used. The clicks are presented at 10-70 times per second. The width of pulse is
generally 100 µs. A series of 5 or more brief peaks is seen in the 6-8 ms after each click
presentation. Since these peaks are less than 1 µV in amplitude, the clicks must be
presented several thousands times to average all backround activity. The classical BAEP
peaks are labelled with Roman numerals I-V, see Figure 38 [Nuwer 1998].

EEG and VEP signal processing 37

5. Evoked Potentials

Figure 38. The principal BAEP peaks are identified by numerals I-V. The peaks seen
here are typical for an adult patient.
Factors influencing peak latencies of BAER include age, sex, auditory acuity
stimulus repetition rate, intensity, and polarity. Rarefaction (earphone diaphragm moves
away from the eardrum) produces increase in wave I amplitude. In severe hearing loss,
all waveforms may be delayed or wave I may be absent with waves II through V
delayed, or all waveforms may be absent. Note that in hearing loss one can still obtain
BAER to assess central conduction time by increasing stimulation intensity.
BAEPs are useful in estimating or aiding in the assessment of hearing loss. The
most commonly used method for this is Evoked Response Audiometry. The frequency
of stimulation is 50-70Hz and at least three different intensities should be used. Wave
five latency shifts are used to estimate the amount of hearing loss.
At this stage, BAEP signals were not recorded nor analysed in the biolaboratory.
We leave the problem for the future work.

EEG and VEP signal processing 38

6. Conclusions

6. Conclusions
In these report we showed the applications of several methods to the analysis of
different type of EEG signals.
The methods such as Fourier transform, orthogonal wavelet parameterization,
Independet Component Analysis have several advantages: low computational
complexity, stable and fast numerical implementations, providing information on the
time, resp. time-frequency distribution of signal’s energy and straightforward
interpretation of coefficients in terms of time-frequency energy content. The
coefficients can be treated as feature vectors ready for statistical evaluation and as well
as the input for the classifier in the iHCI context.
In addition with the brain mapping visualisation methods the results provides a
valuable contribution in developing the Intelligent Human Machine Interface system.
Their importance lie in the parameterisation of psychosomatic quantities as is depicted
in appendix II.
However, despite all advances due to the development of new techniques and
experiments, is still very little what we know about EEG behaviour and underlying
process in the brain.
The future direction of the present study is summarized as follows.
There is enough rooms for others development and application in iHCI
environment above mentioned algorithms. Two topics were not practically covered
during overall processing: measurement and implementation of both the brainstorm-
auditory evoked potentials and the ICA methods. The latter is promising tool for feature
extraction under the iHCI scheme.

This research work has been supported by the internal research program 30-21086-333 of
Czech Technical University in Prague.

EEG and VEP signal processing 39

/. References

7. References
[Achermann et al. 1994] Achermann, P., Hartmann, R., Gunziger, A., and Guggenbühl,
A. All night sleep and artificial stochastic control signals have similar correlation
dimension. Electroencephalography Clinical Neurophysiology, 90:384-387, 1994

[Bates_WWW] Timothy Bates, Sydney Macquarie University, Depatment of


[Basar 1989] Basar, E. and Bullock, T. H., editors. Brain Dynamics, Progress and
Perspectives, Springer-Verlag, 1989

[Blinowska 1991] Blinowska, K. J. and Malinowski, M., Non linear and linear
forecasting of the EEG time series. Biological Cybernetics, 66:159-165, 1991.

[Burrus 1997] Burrus S., Introduction to Wavelets and Wavelet Transforms, Prentice
Hall, 1997

[Comon 1994] P. Comon, “Independent component analysis, a new concept?,” Signal

Processing, vol. 36, pp. 287-314, 1994.

[Daubechies 1992] Daubechies I., Ten lectures on wavelets, Society for Industrial and
Applied Mathematics (SIAM), 1992

[Dietsch 1932] Dietsch, G. (1932). Fourier-analyse von Elektrenkephalogrammen des

Menschen. Pflüger’s Arch. Ges. Physiol., 230:106-112.

[Durka 1996] Durka, P. J. (1996). Time-frequency analyses of EEG. PhD thesis,

Warsaw University.

[Durka 1995] Durka, P. J. and Blinowska, K. J. (1995).

Analysis of EEG transients by means of matching pursuit. Annals of Biomedical
Engineering, 23:608-611.

[Dvorák 1991] Dvorák, I. and Holden, A. V., editors Mathematical Approaches to

brain funtioning diagnostics. Proceedings in Nonlinear Science. Manchester University
Press, 1991

[Eck 2000] Janku, L. - Šorf, M. - Eck, V. - Lhotská, L. - Fejtová, M. Multiple Criteria

Choice Mechanism for Intelligent Man-Machine Interface. In: Proceedings of INES
Vol. 1. IEEE Industrial Electronics Society., p. 295-298. - ISBN 961-6303-23-6, 2000

[Fisk_WWW] Simon Fisk, University of York, Department of Electronics

[Freeman 1987] Skarda, C. A. and Freeman, W. J. How brain makes chaos in order to
make sense of the world. Behavioral and Brain Sciences, 10:161-195. 1987



EEG and VEP signal processing 40

/. References

[Janku 2000] Janku, L. - Šorf, M. - Eck, V. Preliminary Phase of Co-operation between

Human Operator and Intelligent Interface-Computational Personality Type
Classification. Kesearch Report BIO333-09/00. Praha, CVUT FEL, 38 p. 2000

[Jung 2001], Jung, Tzyy-Ping, Makeig, Scott, McKeown, Martin J., Bell, Anthony J.,
Lee, Te-Won, Sejnowski, Terrence J., Imaging brain dynamics using Independent
Component Analysis , IEEE Proceedings, in press, 2001.

[Lesilie 2001] Leslie Huszar, Department of Neurology, Indian River Memorial


[Lhotska 2000] Šorf, M. - Eck, V. - Janku, L. - Lhotská, L. Application of Neural

Network for Stress Classification. In: The State of the Art in ComputationalI
ntelligence. Vol. 1. Heidelberg: Physica-Verlag. p. 373-375., 2000

[Lopes Da Silva 1996] Lopes Da Silva, F. H. The generation of electric and magnetic
signals of the brain by local networks. In Greger, R. and Windhorst, U., editors,
Comprehensive Human Physiology, volume 1, chapter 25, pages 509-528. Springer-
Verlag. 1996

[Louis_WWW] EEG COURSE and GLOSSARY, Department of Neurology, Brown

University School of Medicine

[Louis_WWW 01] Sydney Louis, Department of Neurology, Brown University School

of Medicine,

[Mallat 1993] Mallat, S. and Zhang, Z. Matching pursuit with time-frequency

dictionaries. IEEE Transactions on Signal Processing, 41:3397-3415. 1993

[Maurer 1991] Konrad Maurer, Thomas Dierks, Springer-Verlag, Atlas of Brain-

Mapping: Topographic Mapping of EEG and Evoked Potentials, 1991

[Nuwer 1998] Fundementals of evoked potentials and common clinical applications

today. Electroencephalography and clinical Neurophysiology 106, 142-148, 1998.

[Peter 95] Peter K. H., MD Wong, Lippincott-Raven, Digital EEG in Clinical Practice,

[Pijn et al. 1991] Pijn, J. P., Neerven, J. V., Noest, A., and Lopes da Silva, F. H.
Chaos or noise in EEG signals: dependence on state and brain site.
Electroencephalography Clinical Neurophysiology, 79:371-381, 1991

[Polikar_WWW] Polikar R., The wavelet tutorial

[Renato_WWW] Renato M.E. Sabbatini, PhD Mapping the brain,

EEG and VEP signal processing 41

/. References

[Quiroga 2000] Quian Quiroga R. Obtaining single stimulus evoked potentials with
Wavelet Denoising. Physica D, 145: 278-292, 2000

[Scott 1996] Scott Makeig, Anthony J. Bell, Tzyy-Ping Jung and Terrence J. Sejnowski,
In: D. Touretzky, M. Mozer and M. Hasselmo (Eds). Independent Component Analysis
of Electroencephalographic Data Advances in Neural Information Processing Systems
8, 145-151, 1996.

[Sorf 2001] Five Approaches to Computational Personality Type Classification Problem

CCC 2001-International Carpathian Control Conference, Ladislava Janku, Milan Šorf,
Daniel Novák, Lenka Lhotská, Vladimír Eck, 2001

[Sorf 2000] Šorf, M. - Janku, L. - Lhotská, L. - Eck, V. Applications of Expert System

and Machine Learning Approach to Intelligent Man-machine Interface. In: Intelligent
Techniques and Soft Computing in Nuclear Science and Engineering. Vol. 1. Singapore:
World Scientific. p. 191-197, 2000

[Williams, 1997] Williams, W. J. Recent advances in time-frequency representations:

Some theoretical foundations. In Akay, M., editor, Time Frequency and Wavelets in
Biomedical Signal Processing, IEEE Press Series in Biomedical Engineering. IEEE

EEG and VEP signal processing 42

Appendix I

Appendix I
The International 10-20 System
The International 10-20 System of Electrode Placement (see Figure 39 and 40)
is the most widely used method to describe the location of scalp electrodes. The 10-20
system is based on the relationship between the location of an electrode and the
underlying area of cerebral cortex. Each site has a letter (to identify the lobe) and a
number or another letter to identify the hemisphere location [Fisk_WWW].

Figure 39. International 10-20 system

The “10” and “”20” (10-20 system) refer to the 10% and 20% interelectrode
distance. The letters used are: “F” - Frontal lobe, “T” - Temporal lobe , “C” - Central
lobe , “P” - Parietal lobe, “O” – Occipital lobe.

There is no central lobe in the cerebral cortex. “C” is just used for identification
purposes only.) Even numbers (2, 4, 6, 8) refer to the right hemisphere and odd numbers
(1, 3, 5, 7) refer to the left hemisphere. “Z” refers to an electrode placed on the midline.
The smaller the number, the closer the position to the midline. “Fp” stands for Front
polar. “Nasion” is the point between the forehead and nose. “Inion” is the bump at the
back of the skull.

EEG and VEP signal processing 43

Appendix I

Figure 40. International 10-20 system as viewed from above and from the left

When recording a more detailed EEG with more electrodes, extra electrodes are
added utilising the spaces in-between the existing 10-20 system. This new electrode-
naming-system is more complicated giving rise to the Modified Combinatorial
Nomenclature (MCN). This MCN system uses 1, 3, 5, 7, 9 for the left hemisphere
which represents 10%, 20%, 30%, 40%, 50% of the inion-to-nasion distance
respectively. 2, 4, 6, 8, 10 are used to represent the right hemisphere. The introduction
of extra letters allows the naming of extra electrode sites (see Figure 41).
(Note: These new letters do not necessarily refer to an area on the underlying cerebral

Figure 41. Extended 10-20 system-Modified Combinatorial Nomenclature

EEG and VEP signal processing 44

Appendix II

Appendix II
Intelligent Human-Machine Interface
The relationship between machine and human has been discussed extensively in
terms of the Man-Machine Interface (MMI) and the importance of the MMI has been
recognised in variety of engineering domains. The nuclear engineering and aerospace
engineering are typical examples of such domains, in which human operators are
required to operate huge and complex system through the MMI.
The Human-Computer Interface (HCI), which is one form of MMI, is another
domain with the increasing needs for MMI improvement. The common complaint about
the computer system is that the growing hardware performance does not lead to
friendliness for the users. In the HCI domain, the Graphical User Interface is now
extensively adopted which tries to improve the usability of the computer system.
General description of Intelligent Human-Computer Interface (iHCI)
The interface between man and machine is very actual in the time of computer
industrialisation and also very popular among psychological scientists. The problem of
iHCI belongs under cognitive engineering that is trying to solve questions of mutual
interaction, coexistence between human and machine, performance and workload of
human. Man had to adopt himself to a machine ten and more years ago, the present
approach is exactly the opposite. The designers of complex systems such nuclear and air
engineering try to meet the needs of human in the terms of best performance level with
minimisation of errors occurrence during working process of man. In the subsequent
section we will briefly summarise the concept of iHCI.
Concept of iHCI
The important feature of iHCI concept is schematically illustrated in Figure 42,
and it is composed of the following six basic functional modules.
i. Psychosomatic quantities measurement – measurement of physiological
parameters such as EEG, ECG, SPR, EMG.
ii. Parameterisation of Psychosomatic Quantities – parameterisation of biological
signals by means of signal processing theory.
iii. Estimation of Psychosomatic States - estimation of the cognitive state (tested
person is under non-stress, low-stress, high-stress, etc.), workload, performance
level, error occurrence.
iv. Test Selection – some mental task presented to the tested person to evoke stress.
v. Models – prediction of human behaviour,
vi. Feedback Controller – control of information flow to the screen.

EEG and VEP signal processing 45

Appendix II

Human state estimation by physiological means

The fundamental idea how to estimated human state is based on the measuring
of dynamic changes of various psychological measures such as EEG, ECG, EMG, etc.
by means of medical equipment such as polygraph, and then processing the
characteristic (parameterization) of these measured signals to estimate various kinds of
human state such as temporal level of mental workload. Naturally, the EEG signal
involves the biggest amount of information considering human behavior.

Psychosomatic States
Parametrization of


Estimation of
Test Selection




Figure 42. Human state estimation using iHCI

EEG and VEP signal processing 46