bci2000.org BBS

Dear all,

we made different trials of cursor moving task (1D/2D/3D with different number of targets), always following your tutorial step by step.
Nevertheless, we observed that in all our trials the cursor began its movement independently from the user's intention (it seemed it had a movement "by chance"). In the latest trial we would like to have an horizontal movement of the cursor (from left to the right) using C4 as unique input channel (left hand had the best r^2 value) and applying a small Laplacian spatial filter and a "full matrix"(1row and 5 columns in the "spatial filter matrix").
Our parameters file for 1D feedback task with 1 target has the following data:
- in the "linear classifier matrix":

C4_OUT | 11Hz | 1 | 1 |

- in the normalizer section, following the tutorial "trial based 1D feedback task", we set these values:
normalizer offsets = -1
normalizer gains = 2
adaptation = 1

-in the "buffer condition matrix":

(Feedback==1)&&(TargetCode==1)

How can avoid this unvoluntary cursor movements? Did we make any error in one of these parameters?
A last, hopefully easier question: despite this problem, in all the trials we had a result such as "100%correct" and bits transferred were between 0 and 12.
Why we ALWAYS have "100%correct", even if we didn't reach the target, and we didn't transfer any bit?
We read the tutorial, but we couldn't find in it any information regarding this problems. Hopefully, you could help us in the best way!

Thank you so much in advance,

Best regards.

Hello,
There are several points to be made here.

1.

we observed that in all our trials the cursor began its movement independently from the user's intention"

There are two issues leading to this behavior. First, if you are only using a single control input feature (i.e., ch C4, 11 Hz), then the adaptation will very quickly cause the cursor to move in the opposite direction. That is, when the subject is not moving or thinking about moving for a particular target, the adaptation will essentially classify this as the "on" signal for that target. Therefore, right hand activity will move the cursor right, while resting will move it left. The 2nd issue is that there is a high degree of random activity in the EEG; if you get an r^2 of 0.1, then 90% of the EEG activity is NOT related to the task. Therefore, there will likely be a lot of random cursor movement, seeming unrelated to what the user is attempting to do. This will improve with practice, though.

2.

in the normalizer section, following the tutorial "trial based 1D feedback task", we set these values

The normalizer values will change very quickly if adaptation is turned on. The offsets and gains are the values that are normalized during adaptation.

3.

in all the trials we had a result such as "100%correct"

I believe this is the behavior if the trial times out without hitting the correct target. The hit % is only based on the VALID trials, i.e., trials that did not time out. So, if you have 12 trials, and only hit the target once, you will have 100%. I think the only way to get less than 100% is if you turn on the TestAllTargets parameter; otherwise, in the summary window, just divide the "hits" by the "trials" given.