INTRODUCTION

A challenge of BCI research is to be able to control complex devices using noninvasive recordings
of
A challenge of BCI research is to be able to control complex devices using noninvasive recordings
of
 A challenge of BCI research is to be able to control complex devices using noninvasive
recordings of brain signals at high spatial and temporal resolution.
 The cerebral cortex is the area of greatest interest in BCI as
1. It is most accessible to electrode probes and scalp recording
2. It is highly involved in the executive function of motor and communication behaviours.
 The non-invasive recording of brain signals captures the changes in blood flow or
fluctuations in electric/magnetic fields caused by the activity of large populations of
neurons.
 Since EEG signals reflect the combined input to large populations of neurons, methods for
building BCIs from EEG signals rely on modulating the response of large neural
populations either through subject training over a period of time or through external
stimuli that can activate large populations of neurons.

TWO METHODS FOR BUILDING BCIS FROM EEG SIGNALS

1. Self-paced (or asynchronous) BCI

 BCIs based on the subject can voluntarily initiate control at any time without being tied to a
stimulus.
 Self-paced BCIs typically utilize some form of imagery (motor or cognitive) that can
generate a robust and reliable EEG response after a period of training.

2. Stimulus-based BCIs (also called synchronous BCIs)

 Stimulus-based BCIs (also called synchronous BCIs) rely on detecting a stereotypical
brainresponse generated after the subject is presented with a stimulus (such as a flash)
that is linked to a BCI command or choice.
 Thus the Control is not initiated by the subject but is tied to the presentation of stimuli by
the BCI.
 Stimulus based BCIs are however are easier to use because they do not require training on
the part of the subject, and relatively high accuracies can be obtained for naïve subjects,
compared to imagery-based BCIs.
ALL ELECTROPHYSIOLOGICAL SOURCES FOR BUILDING BCIS FROM EEG
SIGNALS
a. Self-paced (or asynchronous) BCI

Self paced BCIs voluntarily initiate control at any time without being tied to a stimulus. They
typically utilize some form of imagery (motor or cognitive) that can generate a robust and
reliable EEG response after a period of training.

 Oscillatory Potentials and ERD event-related desynchronization motor imagery-based

BCIs, where users modulate their brain activity to produce changes in oscillatory power.
 Slow Cortical Potentials SCPs are type of electrical brain activity characterized by slow
changes in cortical voltage over time, typically lasting from several hundred milliseconds
to several seconds. These potentials reflect changes in the excitability of the cortical
neurons and are thought to be involved in various cognitive processes.
 Movement-Related Potentials Movement-related potentials (MRPs) are a type of
event-related potential (ERP) that occurs in the brain before, during, or after the
execution of voluntary movements. These potentials reflect the neural activity
associated with motor planning, execution, and feedback processes.

b. Stimulus-based BCIs (also called synchronous BCIs)

A major class of EEG signals used in non-invasive BCIs are evoked potentials (EPs),
which are stereotypical EEG responses generated by the brain when the sub ject is
presented with a particular type of stimulus.
 The P300 Potential: when a rare but task- relevant auditory, visual or
somatosensory stimulus is interspersed with fre quent and routine stimuli, the rare
stimulus evokes a potential with a positive peak at about 300 ms after the stimulus is
presented. This potential is called the P300 (or P3) potential.
 Visually Evoked Potentials (VEPs): VEPs are generated by visual stimuli such as
flashing lights.
 Auditory Evoked Potentials (AEPs): AEPs are generated by auditory stimuli such
as clicks and tones.
 Steady-State Visual Evoked Potential (SSVEP): SSVEP BCIs use visual stimuli
flickering at different frequencies. The user focuses on a specific flickering
stimulus, eliciting a frequency-specific brain response that can be detected and
used for control.
 Somatosensory Evoked Potentials (SSEPs): These BCIs use tactile or
proprioceptive stimuli to evoke brain responses. For example, vibrotactile stimuli
applied to the skin can elicit event-related desynchronization (ERD) or event-
related synchronization (ERS) patterns that can be used for control.

c. Hybrid BCIs: These BCIs combine multiple types of stimuli or brain signals to improve
performance and usability. For example, a hybrid BCI may combine SSVEP and motor
imagery signals for more robust control.

SENSORIMOTOR RHYTHMS (SMRS)

 In relaxed awake people, the EEG recorded over primary sensorimotor cortical areas often
displays 8–12 Hz (mu-rhythm) and 18–26 Hz (beta-rhythm) activity. These
sensorimotor rhythms (SMRs) are thought to be produced by thalamocortical circuits.
 The execution or imagination of limb movement induces changes in rhythmic activity
recorded over sensorimotor cortex.
 These changes in SMRs can be detected on the scalp by electroencephalography
(EEG) or magnetoencephalography (MEG) or on the surface of the brain by
electrocorticography (ECoG).
 SMRs are oscillations in the electric or magnetic fields recorded over sensorimotor
cortices. (ie) Sensorimotor rhythms (SMR) refer to oscillations in brain activity recorded
from somatosensory and motor areas.
 SMR activity comprises a variety of different rhythms that are distinguished from each other
by location, frequency, and/or relationship to concurrent sensory input or motor
output. Some beta rhythms are harmonics of mu rhythms, while others are separable
from mu rhythms by topography and/or timing.

 Movement or preparation for movement is usually accompanied by SMR decrease,

especially contralateral to the movement. This decrease, known as event-related
 desynchronization (ERD). It consists of a reduction in rhythmic activity related to an
internally or externally paced event such as a voluntary movement. SMRs decrease during
motor behaviors. ERD is also referred to as blocking of a rhythm. SMR ERDs can be seen
as correlates of activated cortical networks.
 SMRs can also be increased in association with sensorimotor events (e.g., immediately
after movement). This is called event-related synchronization (ERS) Correspondingly, the
increase of oscillatory activity in a specific frequency band is called event-related
synchronization (ERS). ERS can be seen as correlate of a deactivated or inhibited cortical
network.
 Furthermore, and most pertinent to BCI use, ERD and ERS do not require actual
movement, they occur also with motor imagery (i.e., imagined movement).
 SMRs typically fall into three major frequency bands: mu (8–12 Hz), beta (18–30 Hz), and
gamma (30–200+ Hz).
 EEG recording is largely limited to mu, beta and lower-frequency gamma activity, but ECoG
and MEG can detect higher-frequency gamma activity.
 Alpha activity recorded from sensorimotor areas is also called mu activity.
 Several laboratories have shown that people can learn to control mu or beta rhythm
amplitudes in the absence of movement or sensation.
 People with or without motor disabilities learn to control mu- and/or beta-rhythm amplitudes.

 Users initially employ motor imagery to control the cursor, but, as training proceeds,
imagery usually becomes less important, and users learn to move the cursor automatically.
  SMR-based BCI has been used to answer simple yes/no questions with accuracies >95%,
to perform basic word processing.
 The frequency bands that are most important for motor imagery are mu and beta in EEG
signals.
 Activity invoked by right hand movement imagery is most prominent over electrode location
C3. Left hand movement imagery produces activity most prominent over C4.That is, activity
invoked by hand movement imagery is located on the contralateral (opposite) side.
 Foot movement imagery invokes activity over Cz.
 A distinction between left and right foot movement is not possible in EEG because the
corresponding cortical areas are too close. Similarly, ERD/ERS patterns of individual
fingers cannot be discriminated in EEG.
 To produce patterns that can be detected, the cortical areas involved have to be large
enough so that the resulting activity is sufficiently prominent compared to the remaining
EEG (background EEG).
 Hand areas, foot areas, and the tongue area are comparatively large and
topographically different. Therefore, BCIs have been controlled by imagining moving the
left hand, right hand, feet, and tongue.
 When there is no movement, the spectrum at C4, which corresponds to the left hand, is
displayed in the following image.

 An imagined movement in left hand causes the SMR rhythm to decrease which is depicted
below.

Mu RHYTHM
 The sensorimotor mu rhythm, also known as mu wave, comb or wicket rhythms or
arciform rhythms, are synchronized patterns of electrical activity involving large numbers
of neurons, probably of the pyramidal type, in the part of the brain that controls voluntary
movement.
 These patterns as measured
by electroencephalography (EEG), magnetoencephalography (MEG),or electrocorticograph
y (ECoG), repeat at a frequency of 7.5–12.5 (and primarily 9–11) Hz, and are most
prominent when the body is physically at rest.
  Unlike the alpha wave, which occurs at a similar frequency over the resting visual cortex at
the back of the scalp, the mu rhythm is found over the motor cortex, in a band
approximately from ear to ear.
 People suppress mu rhythms when they perform motor actions or, with practice, when they
visualize performing motor actions.
 This suppression is called desynchronization of the wave because EEG wave forms are
caused by large numbers of neurons firing in synchrony.
 The mu rhythm is even suppressed when one observes another person performing a motor
action or an abstract motion with biological characteristics.
 The mu rhythm and the beta rhythm are originated from the sensorimotor cortex, i.e., the
area which is primarily responsible for the control of hand and foot movements.
 Alpha waves (or the alpha rhythm) are electrical fluctuations in the range 8–13 Hz and can
be measured in EEG from the occipital region in awake persons when they are relaxed or
their eyes are closed.
 A particular kind of alpha wave popular in BCI applications is known as the mu rhythm (8–
12 Hz).
 Mu rhythm is found over sensorimotor areas in the absence of movement and is decreased
or abolished when the subject performs a movement or imagines performing a movement.
 Mu rhythms display two distinct ERD patterns. Lower frequency (8–10 Hz) mu-rhythm ERD
occurs during almost any kind of motor behaviour, is widespread over the entire
sensorimotor cortex, and probably reflects general motor preparation and attentional
processes.
 In contrast, higher-frequency (10–13 Hz) mu-rhythm ERD is topographically restricted and
is related to task-specific aspects of performance.
 In sum, lower-frequency mu ERD appears to be nonspecific, whereas higher-frequency mu
ERD is topographically and functionally specific.

Slow Cortical Potentials(SCPs)

The slowest features of the scalp-recorded EEG yet used in BCI systems are slow voltage
changes generated in cortex. These potential shifts occur over 0.5–10.0 s and are called slow
cortical potentials (SCPs).

SMRs are EEG features that are analysed in the frequency domain (e.g., frequency on the x-axis)
and thus are often referred to as frequency-domain activity. Additional EEG features associated
with motor function are measured in the time domain and thus are referred to as time-domain
activity.

Movement or movement imagery is typically associated with relatively slow changes in the
voltages recorded over sensorimotor cortex in time domain. These are called slow cortical
potentials (SCPs).

SCPs are event-related potentials that are time-locked and phase-locked to specific sensorimotor
events (i.e., they occur at predictable times before, during, or after specific events).
SCPs typically consist of negative potential shifts that precede actual or imagined
movement or cognitive tasks. They are thought to represent cortical activation in preparation for
action. An SCP is typically followed by a biphasic wave referred to as the movement-related
potential.

In normal brain function, negative SCPs accompany mental preparation, while positive SCPs
probably accompany mental inhibition.

Negative and positive SCPs probably reflect an increase and decrease, respectively, in
excitation of cortical neurons.

The contingent negative variation is a slowly developing negative SCP that occurs between a
warning stimulus and a stimulus requiring a response. It reflects increased activation and
decreased firing thresholds in cortical networks associated with the preparatory process preceding
the response.

Like SMRs, SCPs and related potentials over sensorimotor areas are associated with motor
imagery as well as with actual movements.

They are thought to reflect a mechanism for local mobilization of excitation or inhibition in cortical
populations, caused by inputs from the thalamus.

In one of the studies with 13 healthy subjects and 3 patients with total motor paralysis,
EEG was recorded from electrode locations Cz, C3, and C4 , and two channels were
extracted: a Cz-linked mastoid channel (i.e., 1/2 [(Cz-A1) + (Cz-A2)]) and a bipolar C3 minus C4
channel.

The training task involved controlling a cursor to hit the top or bottom edge of the screen. The
position of the cursor was proportional to the difference between average baseline EEG
amplitude and the average EEG amplitude over the last 500 ms from the Cz channel.

The baseline amplitude was calculated from an immediately preceding baseline period.

Some subjects participated in a two-dimensional cursor task where the target could also be the left
or right edge of the screen. In this case, the horizontal position of the cursor was
proportional to the difference between average baseline EEG amplitude from the (C3−C4)
channel and the average EEG amplitude from this channel over the last 500 ms.
Figure 10 shows the average SCP waveforms generated by healthy subjects on cue after training.
For both channels, a clear deviation from baseline activity in the positive or negative direction can
be seen: this difference from baseline was used to proportionally move the cursor up/down or
left/right. Out of the 13 subjects, 4 were able to produce significant positive-going
responses, 3 generated significant negative-going responses, and 3 were able to generate both.
Slow cortical potentials (SCPs) are slow, mainly negative, voltage shifts recorded over
sensorimotor or frontal cortical areas.

They precede and coincide with imagined or actual motor actions or cognitive tasks.

With extensive training, people can learn to control SCPs and use them to operate spelling
programs and other applications. Although SCP-based BCIs were used successfully in the past by
people with severe disabilities (e.g., Amyotrophic lateral sclerosis (ALS) patent), this BCI modality
receives little attention at present because it is inherently slow, allows only one dimension of
control, requires extensive training, and is prone to error.

In the future, SCP-based BCI paradigms may prove useful as therapeutic neuro-feedback
tools or as adjuncts to other BCI or conventional control modalities (e.g., by allowing recognition
of anticipation in the user).

Distinctive Characteristics of SCPs:

• Frequency Range: SCPs fall in the extremely low-frequency range, typically between 0
and 1 Hz. This is much slower than the alpha, beta, and theta waves commonly used in
EEG (electroencephalography).

• Event-Related: SCPs are not continuous oscillations but rather slow shifts in the baseline
electrical activity of the brain. These shifts are often triggered by external stimuli or internal
events (like mental tasks) and can last for several seconds.

• Shifts in Voltage: SCPs manifest as either negative or positive shifts in voltage relative to
the baseline.

Negative vs. Positive SCPs:

• Negative Shifts: These are believed to reflect increased activation or excitability of the
underlying cortical area. Imagine them as a "go signal" in the brain, promoting processing
or preparation for an upcoming event.

• Positive Shifts: Positive SCPs are thought to indicate inhibition or a decrease in

excitability. They might represent a "hold" or "wait" signal in the brain, promoting a more
relaxed or deactivated state.
SCPs and Neurofeedback:

• SCP neurofeedback is a type of neurofeedback training that focuses on regulating these

slow cortical potentials.

• Users are typically presented with visual or auditory feedback based on their SCP activity.
Their goal is to learn to control these slow shifts in brain activity.

• For example, a user might see a bar graph rise when they produce a negative SCP
(desired activity) and fall when they produce a positive SCP (undesired activity).

Potential Benefits of SCP Neurofeedback:

• Attention Regulation: SCP neurofeedback is being explored for improving attention,

particularly in individuals with ADHD (attention deficit hyperactivity disorder). By learning to
enhance negative SCPs, users might improve their ability to focus and sustain attention.

• Other Applications: Research is ongoing to explore the potential benefits of SCP

neurofeedback for various conditions, including:

– Anxiety

– Depression

– Pain management

– Epilepsy

Potential Applications:

• BCI Control: In theory, SCPs could be used to control external devices by associating
specific mental tasks (reflected in SCP shifts) with desired actions.

• BCI Calibration and User State Monitoring: SCPs might be helpful for calibrating BCIs or
monitoring the user's state of attention or engagement. Negative SCPs could indicate focus
and readiness for BCI interaction, while positive SCPs might suggest drowsiness or
disengagement.

Challenges of Using SCPs in BCI:

• Slow Nature: The sluggish pace of SCPs (0-1 Hz) makes them less ideal for real-time
control compared to faster brainwaves like SMRs (around 12-20 Hz) that offer quicker
signal changes.

• Limited Spatial Resolution: Due to their slow nature, SCPs are typically more difficult to
localize compared to faster brainwaves. This can make it challenging to pinpoint the
specific brain region generating the SCP activity.

• Mental Effort: Effectively modulating SCPs often requires sustained mental effort, which
can lead to fatigue and hinder BCI performance.

MOVEMENT-RELATED POTENTIALS
EEG signals show a small and slow potential drift prior to voluntary movements.
These movement-related potentials (MRPs), sometimes also called readiness potentials
(RPs) or Bereitschaftspotentials (BPs), show variation in distribution over the scalp with respect to
the body part being moved.

The Bereitschaftspotential (or readiness potential) is a negative SCP that usually begins 500–1000
ms before a self-initiated movement.For example, the BP related to movement of left versus right
arm shows a strong lateral asymmetry.

This potentially allows one to not only estimate the intent to move, but also distinguish between left
and right movement intention.

This makes them attractive targets for BCI applications but since they are typically much smaller
than other EEG phenomena such as alpha or beta rhythms, their detection is much harder.

It has been suggested that while ERD may reflect changes in the background oscillatory activity in
wide cortical sensorimotor areas, MRPs may represent increased, task-specific responses of
supplementary and primary motor cortical areas.

Its amplitude and topography are affected by movement type and the muscles involved.

Types of MRPs:

• Bereitschafts potential (BP) or Readiness Potential: This slow, negative shift in the EEG
signal starts about 1.5 seconds before a self-initiated movement. It reflects the brain's
preparation and intention to move.

• Contingent Negative Variation (CNV): This potential arises when a cue or warning signal
precedes a required movement. CNV appears slightly before the BP and reflects
anticipation and preparation for the specific cued movement.

Decoding the Signals:

BCI systems analyze MRPs to understand the user's movement intentions.

• The timing and location (on the scalp) of the MRP can provide clues about the type of
movement being planned (e.g., left arm vs. right leg).

• The amplitude (strength) of the MRP can sometimes indicate the intensity of the intended
movement.

Advantages of MRPs in BCI:

• Early Detection: MRPs appear before the actual movement, allowing BCIs to predict and
potentially assist with the intended movement.

• Works with Imagined Movements: Similar brain activity patterns occur during both actual
and imagined movements. This allows users to control BCIs without physically performing
the movements, which can be helpful for individuals with paralysis.

Challenges and Advancements:

• Lower Signal Strength: MRPs are typically weaker compared to other brain signals like
ERPs (event-related potentials), making them more susceptible to noise and requiring
advanced processing techniques.
 • Variability: The timing and amplitude of MRPs can vary between individuals and across
different movement tasks.

• Limited Control Signals: Decoding complex movements with high precision solely from
MRPs can be challenging.

STIMULUS-EVOKED POTENTIALS
A major class of EEG signals used in noninvasive BCIs are evoked potentials (EPs),
which are stereotypical EEG responses generated by the brain when the subject is presented with
a particular type of stimulus.

For example, when a rare but task- relevant auditory, visual or somatosensory stimulus is
interspersed with frequent and routine stimuli, the rare stimulus evokes a potential with a positive
peak at about 300 ms after the stimulus is presented.

This potential is called the P300 (or P3) potential Other types of responses include:

 Visually evoked potentials (VEPs) generated by visual stimuli such as flashing lights,
 Steady state visually evoked potentials (SSVEP) produced by a visual stimulus repeated
at a rate greater than 5 Hz,
 Auditory evoked potentials (AEPs) generated by auditory stimuli such as clicks and
tones, and
 Somatosensory evoked potentials (SSEPs) caused by somatosensory stimulation.

THE P300 POTENTIAL

The P300 (or P3) signal is so named because it is a positive deflection in the EEG signal that
occurs approximately 300 ms after a stimulus. The stimulus itself must be rare and unpredictable
but relevant to the subject (e.g., sudden intensification of an attended target).

The amplitude of the P300 depends directly on how relevant the stimulus is and varies inversely
with the probability of the stimulus.

The P300 is generally observed most strongly over the parietal lobe, although some components
also originate in the temporal and frontal lobes.

Although it often occurs at a latency of about 300 msec relative to the eliciting stimulus
(hence the designation of P300), its latency may vary from 250 to 750 msec.

This variability in latency reflects the fact that the P300 is elicited by nature of the decision.

The P300 is usually largest over central parietal scalp and attenuates gradually as distance from
this area increases.

The voltages that constitute the ERP are embedded within the general EEG activity recordable
from the scalp and are usually quite small relative to the ongoing EEG.

However, because the ERPs are time-locked to events, and follow a constant time course, they
can be extracted by averaging multiple trials of eliciting events.

The result is a series of positive and negative voltage deflections that are referred to as
components.

Longer-latency components ( > 150 msec) tend to reflect information-processing activity that is
cognitive in nature and is thus less dependent on stimulus modality and more dependent on the
significance of the eliciting event in the subject’s concurrent tasks. They are usually referred to as
endogenous components.

Current P300-based BCIs allow users to select items displayed on a computer screen.

The specific set of circumstances for eliciting the P300 ERP is known as the Oddball Paradigm.
This paradigm has three essential attributes:

 A subject is presented with a series of events (i.e.,stimuli), each of which falls into one of
two classes.
 The events that fall into one of the classes frequent than those that fall into the other class.
 The subject performs a task that requires classifying each event into one of the two classes.

The events that fall into the less-frequent class elicit a P300.

As long as an experimental design adopts the three attributes of the oddball paradigm, any
stimulus and any classification task can elicit a P300.

A P300 can be elicited by an event that consists of the absence of a stimulus, if that absence
satisfies the conditions of the oddball paradigm.

Figure 12 illustrates a typical P300 experiment. The letters ‘O’ and ‘X’ flash on a video screen
in a random order at a rate of one per second.
The X occurs infrequently (e.g., 20 % of the flashes) and is thus the oddball stimulus, while the O
occurs frequently. The subject is asked to count the number of times one of the stimuli (e.g., X )
occurs.

Each time a stimulus occurs, a marker is placed in the data file to indicate the identity of the
stimulus, X or O.

Each stimulus is presented on the screen for 100 msec, and then the screen is blank for 900 msec
until the presentation of the next stimulus.

The ERPs elicited by the oddball stimulus at midline electrode locations Fz, Cz, and Pz of the 10–
20 system for 800 msec after the stimulus.
The three responses show a typical P300 scalp topography: the most prominent potential is a
positive component occurring about 350 msec after the X stimulus; and it is largest at the
Pz electrode and attenuates at more anterior and posterior locations.

It should be noted that the results would be essentially the same even if the subject had been
asked to count the frequent O stimuli rather than the rare X stimuli.

- P300 is always elicited by the rare events.

- P300 latency may vary from 250 to 750 msec.

A famous example of an early BCI based on EEG is the P300 now-classic BCI “speller” based on
the oddball paradigm, the 26 letters of the English alphabet are displayed in the form of a 6 × 6
matrix on a computer screen.

In order to spell a word, the subject must select each letter comprising the word by focusing
attention on that letter in the matrix.

While the subject is focusing on the letter or command, the rows and columns of the matrix are
repeatedly flashed in random order.

Each flash of a row or column lasts 100 ms, and the interval between flashes is fixed at either 500
ms or 125 ms.
Only when the row or column containing a subject’s chosen letter or command is flashed is a large
P300 generated by the subject’s brain.

The subject’s choice of letter or command can thus be inferred by keeping track of which flashed
row and column elicited the largest P300s.

This signal can be detected using a classifier such as linear discriminant analysis (LDA). The
subject’s choice of letter or command can thus be inferred by keeping track of which f lashed row
and column elicited the largest P300s.

STEADY-STATE VISUAL EVOKED POTENTIALS (SSVEP)

Brain processing of a sensory stimulus or another event can produce a time-locked series
of positive-negative deflections in the EEG.

These event-related potential (ERP) components are distinguished by their scalp locations and
latencies.

Earlier components with latencies <100 ms originate largely in primary sensory cortices
and are determined mainly by the properties of the evoking stimulus.

Later ERP components with latencies of 100–>500 msec reflect to a greater extent ongoing brain
processes and thus are more variable in form and latency.

SSVEP stimuli do not flash successively, but flicker continuously with different frequencies in the
range of about 6–30 Hz. Paying attention to one of the flickering stimuli elicits an SSVEP in the
visual cortex that has the same frequency as the target flicker.

That is, if the targeted stimulus flickers at 16 Hz, the resulting SSVEP will also flicker at 16 Hz.
Therefore, an SSVEP BCI can determine which stimulus occupies the user’s attention by looking
for SSVEP activity in the visual cortex at a specific frequency.

The BCI knows the flickering frequencies of all light sources, and when an SSVEP is detected, it
can determine the corresponding light source and its associated command.

Consider a system where the goal is to decode one of two possible choices. One can then
represent the two choices by visual stimuli (e.g., buttons on a screen or light emitting diodes –
LEDs), each blinking at a different frequency.

The subject focuses attention on the button corresponding to his or her choice (e.g., by looking at
it).

This results in an EEG signal in the early visual areas of the brain (the occipital region) oscillating
at the stimulus frequency – this signal is called a steady state visually evoked potential(SSVEP)

By performing a frequency decomposition of the EEG stimulus the BCI can detect the frequency of
the stimulus the user is paying attention to and therefore, the user’s choice.

BCI allowing selection from 13 buttons on a computer screen, representing a virtual telephone
keypad with the digits 0–9, BACKSPACE, ENTER, and an ON/OFF button.
Each of the 13 buttons was flashed on and off at a different frequency between 6Hz and 14 Hz.

To reduce false positives due to alpha rhythms, a screening experiment with eyes closed
was first performed, and frequencies with power more than twice the mean power between 4 Hz
and 35 Hz were excluded from the stimulation frequencies.

Additionally, all stimulation frequencies were odd multiples of the frequency resolution to prevent
one stimulation frequency being twice another stimulation frequency.

In other experiments, the authors found that the minimum difference in flickering frequency
(i.e., frequency resolution) between neighbouring targets that a subject can discriminate is about
0.2 Hz, and the frequency range in which the SSVEP can be effectively observed is approximately
6–24 Hz.

EEG signals were recorded from electrode locations O1 and O2 according to the 10–20 system
with left/right mastoids as reference electrodes.

A fast Fourier transform was performed every 0.3s to compute the amplitude spectrum.

For each stimulation frequency, the sum of its amplitude and that of its second harmonic was used
as the feature for classification.

A simple threshold classifier was used, where the threshold was chosen to be twice the mean
value of the amplitude spectrum between 4 Hz and 35 Hz.The output of the classifier was the
frequency with the largest intensity.

SSVEP-BASED BCI OPERATION AND ANALYSIS

The user views three red boxes each of which flickers at a different frequency. By choosing to
focus on the 8-Hz box, it elicits EEG activity at 8 Hz and its harmonic frequencies (as shown in the
frequency spectrum of EEG activity at occipital location O2; blue trace in bottom panel).
Also shown for comparison is the spectrum produced when the user is not looking at one of the
boxes (red trace).
It lacks the 8-Hz and harmonic peaks but is otherwise similar. The 8-Hz activity is focused over
occipital areas (as shown in the topographical plot [top middle] of power at 8 Hz [with blue
indicating higher power].

The top right panel shows that this increased activity does not occur immediately. The vertical
line indicates the time at which the user decides to look at the 8-Hz box.

Power at 8 Hz and its harmonic frequencies increase significantly (p < 0.01 indicated by blue-
green) over the next 2 sec. (Yellow indicates significant power decrease).
The standard SSVEP BCI paradigm, in which each of the repetitive stimuli occurs at a
specific frequency, is called the frequency-modulated visual evoked potential (f-VEP) BCI
paradigm.

In a t-VEP paradigm, the different stimuli are mutually independent and non-overlapping.

The BCI computes the average VEP for each stimulus and produces the output
represented by the stimulus that elicits the largest VEP.

In a c- (or m-sequence) VEP paradigm, each stimulus occurs in a pseudorandom pattern that is
nearly orthogonal to the patterns of all the other stimuli.

The BCI computes the correlation between the EEG and a template computed for each
stimulus. Typically, the stimulus on which the user is fixating produces the highest correlation.

The accuracy and speed of BCIs that used f-, c-, and t-VEP paradigms are compared.

They found that the c-VEP paradigm performed best, with the f-VEP paradigm yielding
intermediate performance, and the t-VEP BCI much worse.