Biomedical Engineering Reference
In-Depth Information
where
N
is the number of targets and,
P
f
1
(
i
) and
P
f
2
(
i
) are, respectively, the spectrum
peak values of fundamental and second harmonics of
i
th frequency (i.e.,
i
th target)
and
is the optimized weighting factor that varies between subjects. Its empirical
value may be taken as
α
N
1
(
)
()
()
()
=
∑
α =
Pi
Pi
+
Pi
(8.3)
f
1
f
1
f
2
N
i
1
8.2.2.3 Parameter Customization
To address the issue of individual diversity and to improve the subject applicability,
a procedure of parameter customization was conducted before BCI operation. Our
previous study suggests that the crucial system parameters include EEG electrode
location, the visual stimulus frequency band, and the threshold (
T
) for target fre-
quency determination [31]. To maintain the simplicity of operation and efficiency
of parameter selection, a standard procedure was designed to help the system cus-
tomization. It consists of the steps discussed next.
Step 1: Frequency Scan
Twenty-seven frequencies in the range of 6 to 19 Hz (0.5-Hz spacing) were ran-
domly divided into three groups and the 9 frequencies in each group were randomly
assigned to numbers 1 through 9 on the above-mentioned LED number pad. Then
the frequency scan was conducted by presenting the numbers 1 through 9 on the
digitron display one by one and each for 7 seconds. During this time period, the user
was asked to gaze at the LED number pad corresponding to the presented number.
This kind of scan was repeated for three sessions containing all 27 frequencies.
There was a 2-second resting period between each number and a 1-minute resting
period between groups. It took about 8 minutes for a complete frequency scan. The
7-second SSVEP response during each frequency stimulus was saved for the follow-
ing offline analysis. In the procedure of frequency scanning, the bipolar EEG elec-
trodes were placed at Oz (center of the occipital region) and one of its surrounding
sites (3 cm apart on the left or right side). According to our previous study [31, 36],
this electrode configuration was the typical one for most users.
Step 2: Simulation of Online Operation
The saved EEG segments were analyzed using the FFT to find the optimal frequency
band with relatively high
Q
values. The suitable value of the threshold
T
and the
weight coefficients were estimated in a simulation of online BCI operation, in
which the saved EEG data were fed into the algorithm in a stream.
Step 3: Electrode Placement Optimization
Only one bipolar lead was chosen as an input in our system. For some of the sub-
jects, when the first two steps did not provide reasonable performance, an advanced
electrode placement optimization method was employed to find the optimal bipolar
electrodes. The best electrode pair for bipolar recording with the highest SNR was
selected by mapping the EEG signal and noise amplitude over all possible elec-
Search WWH ::
Custom Search