yeeouch! that helmet does not sound very comfortable, though I guess everyone to their taste
I have very similar results with ECG on the ADS1299. I can grab bare leads with my fingers and get a very well defined and noise free (60Hz removed) trace. Chip has also shown that off-the-shelf ground clips and paste can perform very well compared to more 'legit' EEG electrodes. Were those tin plated Chip?
It looks like you are using zinc plated washers in your setup... what kind of galvanic system have you created lol!
I have got great results with the active electrode design from Olimex (OpenEEG), which is no great surprise... I think an active dry contact electrode will be the best fit for usability. Form factor and comfort are another question...
In the OpenBCI design, we use a split supply, and have had some issues with the bias. It seems there's a discrepancy between the datasheet and the dev mod schematic when it comes to the external reference circuit. In the datasheet, VCAP1 is coupled to AGND (it would appear) and in the dev mod schematic VCAP1 is coupled to AVSS... I notice a difference, but I also need to set up a scope and really watch the bias.
Does your tilt-o-meter adjust for noise in the signal? or ignore data that is coincident with movement?
Haha, maybe not the most comfortable but hopefully reproducible. Once the Spider Claw 3000 is out I'll be looking at that ;-). For now hopefully if this hat works I can pad the inside with foam and make it reasonable and quick for on off. I just didn't have a better idea
"Were those tin plated Chip? It looks like you are using zinc plated washers in your setup… what kind of galvanic system have you created lol!"
That's something I'd like to know more about - any pointers about electrode materials and the implication of it? I just used those washers because they were the first thing I found. AgCl disks are pricey!
I agree in the long active electrodes are the way to go. My goal was to get data to play with sooner. As far as I know, no one claims active is better overall, right? Just that it is tons more convenient.
That's interesting about the distinction on AVSS and AGND for the VCAP1 signal. You think it is an error in the data sheet? I can definitely see how the chip would be sensitive to the difference but it seems like the relevant issue will be the power supply noise. If AVSS-AGND was perfectly noise free it wouldn't make a difference.
I also see a fair bit of 60 Hz. It's intriguing since my system is battery powered. I want to go do some tests outside and see if that removes it.
Nothing useful from the accel/gyro/mag chip yet. It was just included as much as anything because it was in the design I started modifying and I figured it might be useful for headtracking.
I have always run into the 60Hz noise. Have to travel around the room and such to 'escape' it, usually wind up finding a quiet place, or just turn on the notch filter that Chip made (sweet filter man!)
The principe is pretty fascinating. I get great results with ECG, and sorta ok results from EEG. Lots of drift and noise, etc. Well worth continuing to pursue!
@peabody124, it's fantastic to see other people home-brewing their own electrodes out of cheap stuff from the hardware store. Warms my heart. Back in November, I did a post describing the little guys that I made. Looks a lot like yours!
BTW, if you did want to buy real electrodes (how boring is that!), you can get "disposable" EEG electrodes for a mere $8 for a set of 10. That's a wicked good deal.
And, yes, you can re-use them quite a bit. I think that they're called "disposable" because you probably can't sterilize them up to hospital standards. If you're just using them on yourself, I think that they'd be great.
Thanks for the electrode links. Just ordered a few sets. Should make an improvement on the washers . Those dry electrodes look pretty sweet. Those pogo pins those are something we used in my lab and IIRC they are crazy expensive. You'll end up probably running 80$ an electrode .
I got onboard impedance monitoring working: Seems to produce fairly sane results. I just can't quite determine why the basline impedance is 12k. There is a 10k resistor on both the SRB1 and CH1 leads so to complete the path it should be going across 20k. However, since the SRB1 is high impedance I'm not sure that's even true or what the return path is. When I remove the connection to the bias pin (which isn't high impedance) it doesn't change anything which is quite strange to me.
However, the delta when I change the resistance is sane so I feel reasonable about it.
Peabody, you are rockin' this thing! Made your own electronics, wrote your own app to drive your test AND to record your data, wrote your own classifier...very impressive.
I'm definitely looking forward to seeing if you can improve the performance of your classifier. If you do, I'd love to see if you can illustrate what signal features / electrode locations contribute heavily to each classifier output state. I do a little machine learning work myself, so I'm definitely interested.
Yeah once I work on electrode placement, focusing on the feature selection is definitely the next big step. I'm in that unfortunate regime where the dimensionality of the data isn't much different than the number of trials. Probably I should also run some experiments that are just left versus right arm for example to really try and power the training. I can't pretend to have have put any real work into any of these issues yet so I really hope there should be lots of room to improve.
James, again, very cool tools. Have to agree with Chip. Umm, with that black swim cap on your head, you do look a bit like a superhero. :-) The "Winged Avenger", perhaps. Thanks for showing us the way here.
James, hi. I'm posting a comment here in the new forum partly to see if you get an email notification. And partly to ask you for an update on your Sparky ADS1299 board. Went over to your UAV site, http://buildandcrash.blogspot.com/ . Cheers, William
As per my experience it is interesting about the distinction on AVSS and AGND for the VCAP1 signal. May be it is an error in the data sheet. I can definitely see how the chip would be sensitive to the difference but it seems like the relevant issue will be the power supply noise. Only if AVSS-AGND was perfectly noise free it wouldn't make a difference.
Comments
I have very similar results with ECG on the ADS1299. I can grab bare leads with my fingers and get a very well defined and noise free (60Hz removed) trace. Chip has also shown that off-the-shelf ground clips and paste can perform very well compared to more 'legit' EEG electrodes. Were those tin plated Chip?
It looks like you are using zinc plated washers in your setup... what kind of galvanic system have you created lol!
I have got great results with the active electrode design from Olimex (OpenEEG), which is no great surprise... I think an active dry contact electrode will be the best fit for usability. Form factor and comfort are another question...
In the OpenBCI design, we use a split supply, and have had some issues with the bias. It seems there's a discrepancy between the datasheet and the dev mod schematic when it comes to the external reference circuit. In the datasheet, VCAP1 is coupled to AGND (it would appear) and in the dev mod schematic VCAP1 is coupled to AVSS... I notice a difference, but I also need to set up a scope and really watch the bias.
Does your tilt-o-meter adjust for noise in the signal? or ignore data that is coincident with movement?
Cheers
"Were those tin plated Chip? It looks like you are using zinc plated washers in your setup… what kind of galvanic system have you created lol!"
That's something I'd like to know more about - any pointers about electrode materials and the implication of it? I just used those washers because they were the first thing I found. AgCl disks are pricey!
I agree in the long active electrodes are the way to go. My goal was to get data to play with sooner. As far as I know, no one claims active is better overall, right? Just that it is tons more convenient.
That's interesting about the distinction on AVSS and AGND for the VCAP1 signal. You think it is an error in the data sheet? I can definitely see how the chip would be sensitive to the difference but it seems like the relevant issue will be the power supply noise. If AVSS-AGND was perfectly noise free it wouldn't make a difference.
I also see a fair bit of 60 Hz. It's intriguing since my system is battery powered. I want to go do some tests outside and see if that removes it.
Nothing useful from the accel/gyro/mag chip yet. It was just included as much as anything because it was in the design I started modifying and I figured it might be useful for headtracking.
I have been messing around with capacitive electrodes. Have you heard of them?
http://www.mdpi.com/1424-8220/11/6/5819/pdf
The principe is pretty fascinating. I get great results with ECG, and sorta ok results from EEG. Lots of drift and noise, etc. Well worth continuing to pursue!
http://eeghacker.blogspot.com/2013/11/homemade-passive-electrodes.html
BTW, if you did want to buy real electrodes (how boring is that!), you can get "disposable" EEG electrodes for a mere $8 for a set of 10. That's a wicked good deal.
http://www.neuro-source.com/product.php?id_product=544
And, yes, you can re-use them quite a bit. I think that they're called "disposable" because you probably can't sterilize them up to hospital standards. If you're just using them on yourself, I think that they'd be great.
Chip
I got onboard impedance monitoring working: Seems to produce fairly sane results. I just can't quite determine why the basline impedance is 12k. There is a 10k resistor on both the SRB1 and CH1 leads so to complete the path it should be going across 20k. However, since the SRB1 is high impedance I'm not sure that's even true or what the return path is. When I remove the connection to the bias pin (which isn't high impedance) it doesn't change anything which is quite strange to me.
However, the delta when I change the resistance is sane so I feel reasonable about it.
you should also check out this thread
http://www.openbci.com/forums/topic/ads1299-electrode-impedance-measurement-algorithm/
I don't think the pogo pins are the way to go for electrodes at all. too expensive, too bulky and uncomfortable...
We had very mixed results from the capacitive prototypes we made. Definately worth more testing, but also could wind up being a 'repeatability' problem...
http://www.nytimes.com/2014/01/21/science/new-truths-that-only-one-can-see.html?hpw&rref=science&_r=0
http://buildandcrash.blogspot.com/2014/03/eeg-and-decoding.html
I'm definitely looking forward to seeing if you can improve the performance of your classifier. If you do, I'd love to see if you can illustrate what signal features / electrode locations contribute heavily to each classifier output state. I do a little machine learning work myself, so I'm definitely interested.
Chip
signal.
May be it is an error in the data sheet.
I can definitely see
how the chip would be sensitive to the difference but it seems like the
relevant issue will be the power supply noise.
Only if AVSS-AGND was
perfectly noise free it wouldn't make a difference.