Information Technology Reference
In-Depth Information
EEG in Fig.
2.5
b. The objective of noise reduction would be to reduce the noise as
much as possible without distorting the signal contents.
The signals with noise reduced are then sent to the feature extraction stage, where
mathematical models such as autoregressive (Huan and Palaniappan
2004
) are used
to extract parameters representative of the signal. Nowadays, nonlinear methods
such as Lypunov and approximate entropy coef
cients (Balli and Palaniappan
2013
)
are also being used to obtain more accurate representation due to EEG signals being
nonlinear. Nevertheless, linear methods are still popular due to their simplicity and
ease of computation.
The extracted features are then classi
ed into respective categories depending on
the application. In some BCI paradigms [such as the SSVEP (Miranda et al.
2011
)],
the classi
ers such as neural network
(Huan and Palaniappan
2004
) and linear discriminant analysis (LDA) (Asensio
et al.
2011
) are used. The
er is relatively simple but in others, classi
final stage is the device control stage where the BCI
output is used to control an external device (for example to select on-screen menus
or move a wheelchair). Certain BCIs employ feedback of the output to improve the
reliability of the system.
2.3
EEG-based BCI Paradigm 1
—
Motor Imagery
Voluntary movement is composed of three phases: planning, execution and
recovery. Even during imaginary movement (known as motor imagery), there is the
planning stage that causes a change in EEG. For example, imagined movements of
left hand causes a change known as event-related desynchronisation (ERD) in the
right motor cortex area, i.e. contralaterally to the imagined movement side and
event-related synchronisation (ERS) in the left motor cortex area. Discrimination of
these ERD/ERS can be used to design a BCI.
2.3.1 ERD/ERS
ERD and ERS generally occur in mu (
20 Hz) fre-
quency ranges. ERD is the EEG attenuation in primary and secondary motor cor-
tices during preparatory stage which peaks at movement onset in the contralateral
hemisphere while ERS is EEG ampli
8
12 Hz) and beta (
13
*
-
*
-
cation in ipsilateral hemisphere occurring
during the same time. ERS appears to be an evolutionary built-in inhibitory
mechanism, which explains why it is dif
cult to execute dissimilar tasks on both
sides of the body simultaneously.
2
In addition to mu and beta frequency ranges, sometimes there is also an increase
in EEG energy in gamma (>30 Hz) frequency range. A simple electrode set-up for
2
This can be demonstrated using an old trick. While sitting comfortably, lift right leg off the
ground and rotate the right foot clockwise. Now, with right hand, draw number six in the air
—
what happens to the foot direction?
