P300 speller predictions with Neurosky are going bad

[Sergio Vivar]

Hi
Im working with a Neurosky headset in P300 speller and the predictions are very bad.
I get the first signal acquisition, I run the train classifier, then I run the online escenario, I run again the train classifier with the file recorded in the online scenario, and another two cycles like that. My predictions is sometimes only “A” and sometimes some other letter but never the right letter.

I place electrode in Pz, I use the escenario provided by Openvibe and i only made this changes:

*Channel Selector= 1 index
*Lua stimulator “Launch” time to send=5

I placed a signal display box after the Temporal Filter in online scenario for showing the signal while I make the exercise and i recorded a video for see it later and I cannot apreciate the P300 wave. I provide a screenshot, to see if you can tell me if the P300 is observed. I olso provide my files for if you can review them

Openvibe versión: 1.2.2
SO: Windows 10 x64
Sampling frequency: 512
Sample count per sent block: 32
Preferences -> Global configuration: Default

In this moment must appear a P300 signal, but i cant view it


Files:
https://mega.nz/#!OcMjTYoQ!t1nxZ6y8JiD ... JHUd_Ezys

Please help me, thanks.

Sergio Vivar

[jtlindgren]

Hi Sergio,

its probably very challenging trying to detect P300 from a single electrode. Typically whats done is that multiple electrodes are integrated in some manner, and then many trials (flashes) are integrated as well, until after enough repetitions, a P300 pattern can be discernible in the stimilus-onset aligned average signal over these flashes. The letter choice also usually depends on combining predictions from many flashes.


Best,
Jussi

[Sergio Vivar]

I´ll try OpenEEG, 2 lectrodes, i want better results with this hardware. Thanks by your answer Jussi.