Information Technology Reference
In-Depth Information
The P300 Composer is like the P300 speller, but instead of displaying a grid of
letters, it displays a grid of possible music note-names from A1 to G5, and addi-
tionally numbers and/or spaces that can indicate rests or other actions. Each grid
element
flashes an equal number of times, but in a random order, and the P300
averaging technique described above is used to detect the grid element having the
highest average amplitude signal when compared against the others. As previously
described, the user must attend to a speci
c grid element, actively noticing when it
flashes. The user can do this either by mentally counting the number of
ashes, or
by some other method they
find works for them.
In order to determine which note-name or rest a user is attending to (i.e. looking
at), each position on the grid needs to
flash enough times for the system to be able
to compute a usable averaged signal representing each possible choice. As already
mentioned, this depends on a number of factors, including the signal condition, the
user
is, skill and experience, how tired the user is, how much they blink or move etc.
The more time the user is able to spend, the more likely they are to be able to
indicate their choice accurately, with the knock-on effect that with 42 different
elements in the grid, the system can be slow to use. For example, if each note
'
flashes 20 times, for a combined on/off duration of 100 ms each time, each note-
name will take 84 s to be detected. However, it will be very accurate.
There are a number of methods for decreasing the time taken to detect ERPs. For
example, each column and row can be
flashed in sequence, which dramatically
speeds up the process. If such a method is used, each XY position on the grid is
found by
flashing each row and column once in random order. The currently
attended area of the screen is detected by comparing the average of all rows (X) and
all columns (Y) to
find the highest average peaks in each, such that they correspond
to the target row and column. With this approach, the time taken to perform each
test drops to around 26 s.
One can again speed up the process by reducing the
flashing speed, and the inter-
stimulus interval. For example, each
flash might be fast as 50 ms (one-twentieth of a
second), with a gap of 20 ms between
flashes. This would reduce the time taken to
detect the correct note-name choice to around 18 s.
Furthermore, the number of trials (
(flashes) can also be reduced. It is possible for
an experienced BCMI user to work with a system which uses 7 trials per grid
element or row/column in order to achieve success between 70 and 80 %. This can
lead to a further improvement in speed of approximately 300 %, bringing the total
test-time per note-name choice to just over 6 s if all speed improvements are
applied.
Given the large number of potential interaction choices available to the user with
this method (42 in the matrix presented in Fig.
3.3
), some of the grid elements can
be used to increase the user interaction. For example, these effectively
'
can be used for setting the current note duration, for setting rests of different
durations, and also for indicating that a previous note choice should be removed.
Additionally, play/stop functionality can be offered.
'
empty slots
