[Cyborg] low budget EEG project update

[Dan Brown]

My current task is to determine if the Neurosky modules can produce
raw EEG output, and it appears that answer is that yes, some of them
can. Details below...

So far, I've examined the Neurosky modules from 2 consumer toys, the
ForceTrainer (FT) and the MindFlex (MF). They have slightly different
modules, perhaps because the FT came out earlier than the MF. The one
inside the FT does not have a model name label, but it appears to
resemble what is called the Neurosky TGEM. The module inside the MF is
labelled as a Neurosky TGAM. For each of these, the first thing I did
was connect wires up to the "T" and "R" pins, and read the packets
coming from the module to the motherboard. The packets did not contain
raw EEG signals, but did contain summary data on "attention" and
"meditation" which the toys used. The TGAM also sent 5 eeg power
spectral density numbers for different frequency ranges. I was not
able to get either board to change its behavior while connected to its
motherboard.

Then I broke apart the connections to the motherboards and tried
again. The TGEM responded to 1-byte commands by changing the kinds of
packets it sent, and I found a command that enables the raw eeg
packets (0x1f, aka 31). This should be coming through at a frequency
of 512 Hz, but that appears to be more than the 9600 baud link could
handle (512 * 5 * 8 > 9600), so I'm going to try and find out if I can
change the baud rate. I was not able to get the TGAM to enable raw
eeg. It seemed to only listen to the last 2 bits of the command byte
that I sent it, and I don't know why that would happen.

Next steps:
Try and change the baud rate that the TGEM is using so that it can
send 512 raw eeg packets per second.
Plan a comparison with a known-good eeg system.

I will be traveling for a while, so won't be around to work on this.
If anyone wants to keep things moving while I'm gone, let me know and
I can give you the materials (or suggestions about where to get them).

Dan



On Thu, Jul 22, 2010 at 10:30 AM, Dan Brown <danbrown at gmail.com> wrote:
> Last night Vicente and I met at Noisebridge. We got some data out of
> the headset and talked about it, and came up with some ideas for what
> to do next. Here's a quick summary and some links to more information
> for those interested.
>
> Data validation is an open issue. At times, the numbers coming from
> the Neurosky module appeared to be good; reacting to closing eyes and
> a few other simple tests. However, they also seemed to generate
> similar responses when hooked up to my hand instead of Vicente's (or
> my) head. Next steps for this issue: Dan will get another Neurosky
> module from a Force Trainer and try to get raw data out of it. Then we
> can look at how the raw signal changes when eyes open/closed, or
> blinking lights. If someone has a working EEG, that would be a good
> benchmark for comparison.
>
> Neurosky does claim to be getting its signals from the brain, and not
> from muscles (as the Emotiv headset appears to). Some notes from a
> recent talk by the director of tech development at Neurosky claim this
> to be true. http://codechix.pbworks.com/f/NeuroSkyEEG.html
>
> The headset is a bit uncomfortable. It's made for kids, and the
> elastic strap pushes the metal electrode into the forehead. It would
> be difficult to wear it to sleep.
>
> There are 2 different Neurosky modules that are used in products. One
> is called the TGAM (ThinkGear ASIC Module), and that is what is inside
> the Mind Flex from mattel and the MindSet from Neurosky. The MindFlex
> does not generate raw data, the MindSet does. I tried to send the TGAM
> from Mind Flex commands over the serial link, but none of the ones I
> tried had any effect on its behavior. I wonder if it's pre-programmed,
> or hard wired (reversibly?) to not produce the raw EEG signal? I
> haven't been able to find any data sheets on this chip. The other type
> of module is called the TGEM (ThinkGear EM module, yeah dumb name),
> and it may be what's inside the Force Trainer. Instead of the ASIC
> chip, it has 2 microprocessors (a PIC and an ATMEGA) and it should
> respond to commands to change its behavior. Both types of modules have
> 5 inputs for the EEG: one for the single electrode, and a pair of
> GND+REF for 2 reference electrodes that are meant to be connected to
> the person's ear(s). They also both have the 4 wires for power and
> serial which I mentioned before. This information comes in part from
> the FCC registration docs:
> https://fjallfoss.fcc.gov/oetcf/eas/reports/ViewExhibitReport.cfm?mode=Exhibits&RequestTimeout=500&calledFromFrame=N&application_id=258224&fcc_id='PIYP2639-09A5T'
> (mindflex)
> https://fjallfoss.fcc.gov/oetcf/eas/reports/ViewExhibitReport.cfm?mode=Exhibits&RequestTimeout=500&calledFromFrame=N&application_id=785615&fcc_id='XG9MS1'
> (force trainer)
>
> Here's a recent video with presentations from Zeo, Neurosky, and Brain
> Fingerprinting Laboratories: http://www.vlab.org/article.html?aid=339
>
> And here's a paper with an interesting way to estimate depth of sleep
> in a continuous manner (not epoch-based, not quantized into "stages",
> not based on R&K, not based on FFT):
> http://www.hsr.nl/bobkemp/papers/2000Microcontinuity.pdf
>
> I will set up another time to meet if/when I have a device that can
> generate raw EEG data. In the mean time, I'd be interested to hear any
> suggestions from the rest of you. thanks!
> Dan
>
>
>
> On Tue, Jul 20, 2010 at 9:37 AM, Dan Brown <danbrown at gmail.com> wrote:
>> Last week I picked up the parts. The mind flex was $35 at a surplus
>> store, and the bluetooth module was $45 (a barebones version exists
>> for $25, but for prototyping I wanted it on a convenient breakout
>> board)
>>
>> Yesterday I tapped the serial interface wires on the neurosky eeg
>> daughter board in the mind flex headset, roughly following steps 2 and
>> 3 from this blog post: http://ericmika.com/itp/brain-hack . I took 4
>> wires (+3.3V, GND, TX, RX) and connected them to the bluetooth module.
>> Sure enough, data flowed from the neurosky. It is what we expected:
>> spectral power bands, meditation, concentration. Although I don't
>> expect it to cooperate, I will try to request it to send raw data,
>> too.
>>
>> Today I'm working on writing a python routine to parse the neurosky
>> ThinkGear packets, and if I have time I will try to get a graphic
>> display working.
>>
>> ** WEDNESDAY, AKA TOMORROW at 6pm: I'd like your help interpreting the
>> data. If you're able to make it to noisebridge between 6pm and 8pm or
>> so, you can put on the headset and see your data and then we can try
>> to figure out how good it is. If anyone has another EEG device to
>> compare, that would be very helpful. Please let me know if you plan to
>> be there.
>>
>> In the near future, I will get one of those other EEG toys, the "Force
>> Trainer", which may have a slightly different neurosky module inside.
>> If it does, it may be able to send out raw EEG data at 512 Hz, as well
>> as the stuff that the mind flex provides. At least, that what the FCC
>> registration documents suggest.
>>
>> Dan
>>
>>
>>>>>> >> While for the Mindflex technically has Bluetooth-band wireless is built
>>>>>> >> in, nobody has yet been able to get much use out of it because by default a
>>>>>> >> significant portion of the useful data is filtered out before it is reached.
>>>>>> >>  The $200 standalone model does have full serial over Bluetooth built in if
>>>>>> >> that's what you mean.
>>>>>> >> As for the electrodes, from what I've seen it would seem that they aren't
>>>>>> >> active (though what I've seen isn't hugely informative on the subject), but
>>>>>> >> the amplification/analysis of the system is entirety done inches away from
>>>>>> >> them anyway, so you might get some of the advantages of an active electrode
>>>>>> >> system.  If you were talking about the same data I'm thinking of, it seems
>>>>>> >> that they may have also been using higher quality electrodes, distinct from
>>>>>> >> the ones in their actual consumer products, as a proof-of-concept for their
>>>>>> >> dry electrode system.
>>
>