SSVEP Demo Setup/testing with Emotiv EPOC

Making & changing box plugins and external apps
Post Reply
razano1
Posts: 18
Joined: Wed Jun 20, 2012 1:56 pm

SSVEP Demo Setup/testing with Emotiv EPOC

Post by razano1 »

Dear Jozef...

I am working on ssvep based application using Openvibe, I am using Emotiv Epoc headset. i am having performance issues and i am trying to overcome them...
I have initially tested ssvep demo application ...
I have few queries still unanswered ...

1- you have mentioned in a thread that
the SSVEP scenarios use the electrode positions which are not on the Emotiv Epoc. In ordes to use them with this headset you will have to modify the scenarios - the channel selectors and the spatial filters.

can you please elaborate it a little more.

2- I have performed a small test using sample ssvep dataset.. after training the classifiers i have used this data set as an input in online shooter scenario scenario rather than using acquisition server.
output is observed using stimulation listener box... and it shows that stimulation sequence is similar to default target sequence in training session (considering data is being acquired using default target sequence) .
but if i try to perform same task with data acquired using emotive Epoc this sequence is not as that accurate, although classifier accuracy is almost same in both cases ...
what could be the possible reason for that,?? which transducer is used to capture sample dataset?? i have checked it many times but result is still same....

3- i explored Flipswitch box lua script but not been able to understand its usage in training and classification scenarios??
What exactly this box do ?? is it only to control termination of scenario or what ??

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

5- What is the impact of Processing Epoch Interval in SSVEP paradigm ??

any help is greatly appreciated..

Cheers

Asim Raza

jlegeny
Posts: 239
Joined: Tue Nov 02, 2010 8:51 am
Location: Mensia Technologies Paris FR
Contact:

Re: SSVEP Demo Setup/testing with Emotiv EPOC

Post by jlegeny »

Hello,

I am sorry for the later reply, most of us left for vacation for the few last weeks.

I will answer your questions here, also those from thread viewtopic.php?t=1081&p=6715#p6715 :

1.) Older SSVEP scenarios have used named electrodes and spatial filters depending on the number of electrodes, this has been changed with CSP. My comment was actually a mistake.

However, CSP trainer might not always work with signal from Emotiv Epoc. This is due to the fact that the Epoc headset does some processing on the signal (common average) which makes the training algorithm to loop forever. This is because each channel becomes a linear combination of all others. This issue can be avoided by removing one channel from the acquired data (any channel will do).

2.) The sample data was acquired by a high-precision EEG (although using fewer channels than Epoc's 14). It is also very important to setup the Epoc headset well, since it does not have many electrodes on "interesting" places for SSVEP: mainly electrodes like O1, O2, Oz, POz and around are missing. The fact that less electrodes are in the appropriate areas men that some of the classification is probably due to muscular artifacts.

3.) The sole purpose of the flipswitch box is to terminate the scenario after all of the training boxes have finished processing. Basically it waits for a stimulation on each entry and once it has received all of them it sends a stimulation on its own output. This is done so the scenario can be easily used when calling openvibe without GUI.

4.) I have briefly tested the SVM classifier. The results were slightly better. The main reason to use the LDA classifier is that it is simpler and thus better suited for examples on which can people learn and build.

5.) Lowering this interval will result in more commands sent per second. This may and may not be desirable depending on the target application. It does not impact the accuracy in any way.

If you have any more questions please ask.

Hope these answers will help.

Cheers

Jozef

razano1
Posts: 18
Joined: Wed Jun 20, 2012 1:56 pm

Re: SSVEP Demo Setup/testing with Emotiv EPOC

Post by razano1 »

Dear Jozef..

thanks for your detailed response...

So it means CSP selects the desired number(Specified in filter dimension field) of channels/electrodes with best quality of signals amongst others ???
I am thinking to use the subset of EPOC electrodes e.g. only using electrodes in occipital / parietal region by putting headset in reverse direction. it might avoid muscular artifacts to some extent i guess...

Is there any relation between Processing Epoch Interval and Sample Count Sent per block in Acquisition Server ???
as lowering Sample count per block results in smother transition of application .... and from your response it seems processing epoch interval does the same.... is it so ???
target application in my case is web based Google street view app where accuracy is rather more desirable than shorter lag ...

i might be asking you more as i have to perform further testing next week.... :)

Best regards

Asim

jlegeny
Posts: 239
Joined: Tue Nov 02, 2010 8:51 am
Location: Mensia Technologies Paris FR
Contact:

Re: SSVEP Demo Setup/testing with Emotiv EPOC

Post by jlegeny »

Hello Asim,

the CSP indeed selects the best combination of electrodes to maximize the detection one pattern in the signal (this is why there is one filter for each frequency). It should remove the unnecessary electrodes itself, but from experience it might be indeed a good idea to remove the electrodes close to eyes.

For your second question: no, there is no link. Sample Count Sent per block is merely an optimization option to balance the size of the packet/number of packets sen through the network.

As for the Processing Epoch interval please look at this image for explanation:

The yellow, pink and blue rectangles represent the Epochs which are used for classification. The distance between them is the Processing Epoch Interval. The accuracy of the classification can be improved by making the window larger (but this will induce lag, which in your case might not be an issue). Decreasing the PEI will result in less commands being sent.

Cheers
Jozef
Attachments
PEI.png
PEI.png (30.08 KiB) Viewed 7692 times

razano1
Posts: 18
Joined: Wed Jun 20, 2012 1:56 pm

Re: SSVEP Demo Setup/testing with Emotiv EPOC

Post by razano1 »

Dear Jozef..

Thanks for your detailed response...
but one thing is still unclear
in your earlier reply you have mentioned
5.) Lowering this interval will result in more commands sent per second.
but is last reply your statement about PEI is
Decreasing the PEI will result in less commands being sent.
can you please clarify it..

although i have made an inference out of your pictorial explanation that decease in PEI will send more commands ...

I will be asking you a little more as i start final testing of system ..

Cheers

Asim

jlegeny
Posts: 239
Joined: Tue Nov 02, 2010 8:51 am
Location: Mensia Technologies Paris FR
Contact:

Re: SSVEP Demo Setup/testing with Emotiv EPOC

Post by jlegeny »

Hello,

My mistake, the word should be "increasing" instead of decreasing, basically the more you wait to send a command the less of them there will be.

Cheers
Jozef

Post Reply