GDF file for SSVEP experiment

[ipyd]

Hi all,

I am a newbie here.
I just found this useful tool for my BCI class project.

Now, I am working on SSVEP-BCI system and I am going to design my own scenario.
Before going to online experiment, I am planning to do offline experiment using an existing SSVEP data.
Does anyone have the captured SSVEP data?

Actually, i found this file: ssvep-record-[2012.05.05-22.05.48].ov in folder: \openvibe\share\openvibe-scenarios\signals
however, since its format is .ov , I am confused on how to use it.

Thanks for your help.
--ipyd

[fabien.lotte]

Hi ipyd,

Welcome to the OpenViBE forum! the ".ov" extension is the OpenViBE extension, that is the extension of files saved using the generic stream writer. As such you can read it using the generic stream reader (http://openvibe.inria.fr/documentation/ ... eader.html). If I am not wrong the file you mentioned was probably saved with the SSVEP scenario (http://openvibe.inria.fr/steady-state-v ... otentials/) that is available in OpenViBE (Jozef, could you please confirm?). This gives you an idea of what the subject had to do when these signals were recorded.
In case you are not conformtable with this file format, you can still convert it to gdf format (using the GDF writer box) for instance, and then open it in Matlab using Biosig.

Hope this helps,

Best regards,
Fabien

[jlegeny]

Hello ipyd,

the ssvep-record-[2012.05.05-22.05.48].ov is a file which was created by the included ssvep scenario when you have launched it. It is indeed a native OpenViBE format like Fabien said.

There is indeed a sample file of SSVEP EEG. It is in \openvibe\share\openvibe-scenarios\signals\bci-ssvep-training.ov

This file is a recording done by the training scenario which is included. It is tagged as follows :

Four labels define the target at which the person was looking after the stimulation was given:

OVTK_Stimulation_Label00 -> center
OVTK_Stimulation_Label01 -> 20Hz
OVTK_Stimulation_Label02 -> 15Hz
OVTK_Stimulation_Label03 -> 12Hz

Two labels define the beginning and end of flickering periods:

OVTK_Stimulation_VisualStimulationStart -> targets begin to flicker
OVTK_Stimulation_VisualStimulationStop -> targets stop flickering

A way to extract only the interesting slices along with lua scripts to sort the stimulations is illustrated in the ssvep classifier training scenario.

If you have any more questions feel free to ask

Cheers
Jozef

[razano1]

Dear Jozef..

I am currently working on improving performance of my ssvep bci app...

i started with checking ssvep .ov stream file...
i have tried EBML stream spy box for it....
and got the following output

[WARNING] At time 0.000 sec <Box algorithm::Generic stream reader> Did not find output for stream 0 of type identifier (0x403488e7, 0x565d70b6) (type name is [Experiment information])
[WARNING] At time 0.000 sec <Box algorithm::Generic stream reader> Did not find output for stream 1 of type identifier (0x5ba36127, 0x195feae1) (type name is [Signal])
[WARNING] At time 0.000 sec <Box algorithm::Generic stream reader> Did not find output for stream 2 of type identifier (0x6f752dd0, 0x082a321e) (type name is [Stimulations])
[WARNING] At time 0.000 sec <Box algorithm::Generic stream reader> Did not find output for stream 3 of type identifier (0x013df452, 0xa3a8879a) (type name is [Channel localisation])
[WARNING] At time 0.000 sec <Box algorithm::Generic stream reader> Output 0 did not find a stream candidate from the input file
[WARNING] At time 0.000 sec <Box algorithm::Generic stream reader> Current configuration has both file stream(s) with no output connector and output connector(s) with no file stream, you probably should check your box configuration

I have been using training data from this file for my online scenario ....
i got suspicious, whether there might be some problem in this data file... i have tried with other files(captured in training session) as well but same result
your views...

Best Regards
Asim

[jlegeny]

Hello Asim,

these errors appear when the streams which were recorded in a file (by the stream writer box) do not correspond to the outputs of the stream reader box.

This most often happens when you simply insert a Generic Stream Reader box into the scenario and select a file. Indeed, the box must be configured first by adding the right outputs to the box. In your case your box should have four outputs (you can add more from the right click menu) and their order should be : Experiment information, Signal, Stimulations and Channel localisation. The easiest way to obtain such a box is to copy and paste it from the CSP training SSVEP scenario (just change the file to whatever you want to read)

Cheers
Jozef

[razano1]

Dear Jozef..

Thanks for you help it is showing log now...
Can I execute CSP and Classifier-Training Scenarios in fast forward mode ?? I mean does it have any impact on performance / processing ??
In training Acquisition scenario "Target Select" VRPN Button Scenario has 7 input stimulation but Training controller sends four distinct (one for non flickering center target and 3 for flickering stimulus) stimulation other than start n stop Experiment, Start n stop Flickering...
Is there any specific purpose for rest 3 inputs....


Best Regards

Asim

[jlegeny]

Hello,

the trainings scenarios (CSP and Classifier training) are made to be run in fast forward mode, the performance will be exactly the same (note that since k-fold tests are randomized each run will yield a slightly different value, but the classifier itself stays identical).

The additional buttons are there only for convenience, as you might wish to create a scenario composed of more than 4 targets.

Cheers
Jozef

[razano1]

Dear Jozef ...

Have you tested this SSVEP demo with SVM (for classification) ???
What exactly is the reason to prefer LDA over SVM in SSVEP ???

Best Regards

[razano1]

Dear Jozef..

During testing SSVEP demo, I have used larger value for process epoch duration (between 1-5 sec) to achieve almost 100 % accuracy for each stimulating frequency during classification...
but in the online scenario, while debugging with stimulation listener, it shows stimulations in log even when the application is minimized ....
there might be some problem with either lua script of ssvep voter in online scenario(which seems OK to me except one thing i.e. to send a stimulation for NO ACTION) or accuracy shown in the classification process might not be real....

any help on this is greatly appreciated....

Best Regards

[jlegeny]

Hello Asim,

the performances should be higher when using a longer period of time for processing.

The accuracies given by the classifier training scenario are given for each classifier separately, which is why the performance gets overestimated.

You could try to make two sessions with the training scenario. You will train the classifiers with the first run and then input the data from the second pass to the online scenario, this might give you more accurate idea of the performance.

Since you are using an Epoc headset it might also be good to use lower frequencies. Which are you using at the moment?

Cheers
Jozef

[razano1]

Dear Jozef ..

I will try your suggestion ...
I am using 8.57;7.5;6.666;6 frequencies, i have tried with the higher frequencies as well but there is not much difference in performance ...
i haven't tried it with any other headset but with Emotiv EPOC there are too many false positives and false negatives .... might be due to inappropriate positioning of electrodes ....

Best Regards

Asim