[FieldTrip] Emotiv and Fieldtrip - filter in the streamer/buffer?

[Ricardo Jorge]

Hi again,

I'm not sure, to be honest!

I would say fieldtrip is the most precise one, but why would it happen
- I mean, how could it be better than the testbench? That's what's
confusing me: it makes no sense that a signal such as EEG has only
integer values indeed.


But there are weirder things:

In the EDF/CSV files from testbench, the first channel (AF3) has only
integer values, the second channel doesn't have only integer values,
but have mean 260 around, the third channel goes back to the integer
values and 4000 mean again, fourth chan back to ~200 mean, non-integer
and it goes on... which makes me thinking that they probably apply
something, either the baseline and rounding, or something more.

In the EDF files from Openvibe, all numbers are integers as
independently of the channel, but variance's different as well. In
fact, estimating it right now, it says that it has variance around 3.

In the MAT file recorded by Fieldtrip, all numbers are non integer but
with an extremely low variance comparing to the others.


So, for some reason all methods are different but to be honest I feel
that fieldtrip's results are correct and so I'm going to give it a try
and use them in my investigation to see if they can confirm what I
expect.

Thanks once again!

Best regards,
Ricardo Jorge

On Fri, Sep 28, 2012 at 10:39 AM, Ricardo Jorge <landpt at gmail.com> wrote:
> Dear Boris,
>
> Sorry for the late response, I tried to reply as earliest as possible.
> Thank for the answer, by the way.
>
> I don't think it's a scaling issue: it shouldn't affect the variance
> that much and Power spectral seem to be different from both ways.
> Absolutely that I can give you additional information; I am starting
> with explaining almost everything I've done:
>
> 1st: I picked up the function ft_realtime_signalviewer in
> "fieldtrip-20120904\realtime\example" and adapted it slightly so I
> could view it in realtime but at the same time, I could see the
> previous 20 seconds. Besides that, I disabled the
> ft_preproc_baselinecorrect(data.trial{1}) step in that function and
> added some steps that allowed me to save the whole record into a file
> at the same time (is there any function in fieldtrip that does that at
> the same time as realtime?). I'm not sure if I can send files through
> this discussion list: tell me if you want to try it. Just for
> visualization issues, I multiplied the signal in the plot by a factor,
> so it could be similar to Testbench visually, but it doesn't matter
> because I'm not changing any data, just the plot.
>
> 2nd: I opened Testbench and OpenVibe. In order to compare the results,
> I saved them into EDF an opened it in matlab. Testbench has an option
> to set an high-pass filter, but it does not alter the recording, only
> the visualization in real time.
>
> 3rd: I did put the emotiv in the table so it could be at the rest.
>
> 4th: I started all 3 of them. They're not be all exactly at the same
> time because I can't press it all at once (but almost) and so I'll
> record during a reasonable time so I can get enough samples.
>
> Results:
>
> 1. There is a weird thing in the testbench/openvibe data, all values
> in AF3 channel (for instance) are integer numbers, whereas Fieldtrip
> are not:
>
> Fieldtrip:
>  1.0e+03 *
>
>    4.030540530271679   4.030428591191215   4.030553469018260
> 4.030684982830449   4.030580735490210   4.030495361159973
>
> Testbench:
> 4029        4030        4031        4031        4032        4032
>
> They're almost the same and it isn't a problem since they didnt start
> exactly at the same time.
> However, analyzing their mean/variance, we have:
>
> Fieldtrip:
>
> Mean:  4.030526515189773e+03
> Variance: 0.011920664040352
>
> Offline (Testbench):
>
> Mean: 4.030526118470382e+03
> Variance:1.641380154961390
>
> Their mean are great but the same cannot be said to their variance: in
> fact it seems that it spreads way too much while offline. Let's see
> their min and max through a sample:
> Fieldtrip: [4.027692307692308e+03 ;  4.032307692307693e+03]
> Testbench: [4025 ; 4037]
>
> I've put a plot in my dropbox folder if you want to check it:
> https://dl.dropbox.com/u/846911/comparision.jpg
>
>
> I've checked if the signalviewer has any filter, but I can't find any,
> so I cannot understand exactly what could be happening.
>
> Their FFT is similar but not exactly the same as well.
>
> What do you think it could be happening?
>
> Thanks in advance,
>
> Best Regards
>
> Ricardo Jorge