No brain signals -> intermittent reference lead, DC offset

edited December 2015 in Electrodes
Well my board arrived a couple of days ago and I have had limited success with it so far.

I connect lower bias pin to my left ear, lower reference pin to my right, and lower channel 1 to Cz using electrode gel.

The board connects to the dongle fine giving a solid red light, and I can see activity in the GUI when I stroke the pins.

It is also detected in Neuromore, and again if i stroke the pins I see activity but otherwise all channels remain flatlined showing -201.34uV, including the one attached to my head. Removing any of the leads makes absolutely no difference to this no show of data, but if i shake the thing or touch the pins I see lines on the screen and plenty of interference if it goes within a couple feet of my laptop.


What am I doing wrong?


Comments

  • wjcroftwjcroft Mount Shasta, CA
    This is the board that Frank @fgallina sold to you, after he was using it successfully, is that correct?

    Have you tried swapping through some of your leads. It's possible you have a bad reference connection or bad channel connection.

    When you connect all 3 leads, channel, bias, and SRB2 together (with clips or a blob of paste) -- that should produce a signal close to 0 uV.

    Since this board was working correctly at one time, it sounds like a connectivity issue. Make sure also that you are not in any strong EMF fields such as from transformers or power lines. Select the proper notch filter (50 or 60 hz) for your region.

    You mention 'gel', are you referring to the Ten20 paste?

  • edited December 2015
    Yes it's Frank's board. Checked all of the leads with my DMM for continuity and it's Ten20 paste. No EMFs, notch correct.

    image

    The reference did not like being on the ear, this is an electrode at Fp1 , with reference at Cz and ground at left ear. The screenshot caught a low voltage, it was more like 7-10 uV.

    I am able to influence the alpha waves so I know it is showing my head data and not just noise. Eye blinks show up fine, but what is odd is that if I unclip the bias cable from my ear it makes no difference to the signal.

    Still am not able to get this kind of data in Neuromore. I literally closed BCI GUI and opened Neuromore after taking this shot. The active channel is now drifting somewhere around -100uV. There seems to be an AC signal in there but a lot of DC drift.

  • wjcroftwjcroft Mount Shasta, CA
    > The reference did not like being on the ear...

    That's very unusual. Standard setup is with reference on one ear and bias on the other. I asked you to try swapping electrode leads, rather than just checking with a meter. Because if the leads are intermittent when some tension is placed on the wires, connectors, or cup attachment, then you may have different results than with just doing a static meter check.

    It looks like you're getting more normal readings now. Put your electrode on O1 O2 or Pz and see if you can pickup the alpha with your eyes closed.

    Another typical location used for the Bias lead (if both ears are linked in "linked ears" configuration) -- is to place the Bias somewhere on the midline, at an unused site. The Electro-Caps put Bias/Ground at AFz. Cz would also be fine. The reason for the midline Bias is so that the power lines cancellation injection hits both sides of the head more evenly.

    > if I unclip the bias cable from my ear it makes no difference

    Try placing it on the midline instead. Bias is part what keeps the differential amplifier "centered" and within limits. If you already had a connection moments ago, then that condition persists for some time. However you should see a reduction in 50 hz level with Bias connected

    You can disable the unused channels so that they are not showing "Railed".

    The ADS1299 and OpenBCI are what is called full band EEG,

    http://openbci.com/forum/index.php?p=/discussion/201/large-millivolt-data-values-fbeeg-full-band-eeg

    In general, when dealing with VPL element blocks, the raw EEG signal must first be subjected to a filter to remove the millivolt DC offset. Typically a high-pass filter at .1 or .5 hz will do this just fine.

    William



  • edited December 2015
    I gave the cables a good wiggle whilst testing and have no reason to suspect that the lead is broken having used it for a couple of days. All I did from previously was to move the reference to my head. I have very little body fat ±5%, might this affect continuity to extremities like earlobes?

    I understand about filtering off the low frequency stuff but in Neuromore raw EEG display isn't anywhere near 0uV Is that ok? I can't see any kind of hardware settings in the app. The signal looks good in the OpenBCI GUI but all I see in neuromore is the offset. and despite having filters in the protocol it is not extracting any features.

    Overall the device seems temperamental. Sometimes it works, sometimes it doesn't so it will take me a little while to get the hang of it. Would it be worthwhile getting some disposable adhesive electrodes, do they ensure a better signal?
  • wjcroftwjcroft Mount Shasta, CA
    Possibly try the BrainBay tutorial example. It has explicit filters in the signal path.

    You have some type of intermittent connection with your reference lead. The standard way to debug that is to swap out with another lead. Body fat should have nothing to do with it. It may be possible that skin surface oils are influencing the electrode cup / skin interface. Try wiping the area with some alcohol first before applying. Use an additional piece of tape to hold it on your ear.

    Were you able to see the alpha with your eyes closed in the GUI? Also try some of the other tests shown in the tutorial doc.

    http://docs.openbci.com/tutorials/01-GettingStarted#getting-started-w-openbci-v-connect-yourself-to-openbci-1-what-you-need

  • Yes I can see lots of alpha at O1, and eyes blinks etc all coming through at forehead sites. Everything seems to respond as it should in the BCI GUI. If I tap the electrode I see spikes, and removing bias or reference causes disturbance.

     Is it ok for the signal to be at -90 in Neuromore? It doesn't look anything like the test data and the magnitudes it is picking up are tiny.

    The leads are OK, I'm experienced with electronics but a newbie to EEG so that is likely the problem.

  • wjcroftwjcroft Mount Shasta, CA
    Once the DC offset is removed, all EEG signals will center around 0 uV, plus or minus at the most say 50 uV. If you are seeing -90 uV DC offset in neuromore, then something is amiss with your highpass filter. Have you tried some of the neuromore provided example classifiers setup for OpenBCI? Also try setting up an extremely simple design with just the OpenBCI connected to a highpass filter.

  • edited December 2015
    Well the filter in Neuromore OSX doesn't seem to work (have been doing this on my Macbook). A spectrum view after the filter shows a huge sub 1hz spike even with the high pass set at 5Hz.

    Considering the GUI is showing data without issue I am going to try again tomorrow with some surgical alcohol and OpenViBE on the NUC instead.

    Thanks very much for your attempts at trouble shooting :)
  • wjcroftwjcroft Mount Shasta, CA
    If you have your NUC setup, then certainly load the BrainBay example and try that before the OpenViBE. The BrainBay example is explicitly geared to neurofeedback.

    It's possible the neuromore highpass might have an issue. You could instead use a Butterworth bandpass from say .5 to 45 hz. The filter order also makes a difference in terms of how sharp the skirts are. BrainBay, BioEra and Bioexplorer will all draw a graph of the filter bandpass characteristics; if you look inside the filter element block. Bioexplorer will also even show you the changes in milliseconds delay over that bandpass.


  • edited December 2015
    Just tried it on the PC and it is exactly the same. Highpass/band pass made no difference. The OpenBCI demo shows nothing on the FFT graphs other than the LF spike.
  • wjcroftwjcroft Mount Shasta, CA
    OK, you're still talking about neuromore, this time on the PC?

    Try the BrainBay tutorial, then if you want to get neuromore to work, play with the filter specs there. Something must be amiss, or they have a glitch in their implementation that you just happened to stumble on.

    https://sites.google.com/site/biofeedbackpages/brainbay-openbci

  • Brainbay works fine. I can see my SMR, theta and beta bobbing up and down as I type.

    The filters work ok and the LF noise is not an issue. Traces are centred on the scopes and it feels like neurofeedback.

    However there's no inbuilt MIDI synth on the NUC and it won't play any of my .avi files :)
  • wjcroftwjcroft Mount Shasta, CA
    Your Windows does not have a General MIDI built in? I thought that was in all Windows. Umm, but see 2nd link, apparently Windows 10 stumbles.


    There is a video player in BrainBay, I've not used it. You may be able to transcode your avi into something it will understand.



  • wjcroftwjcroft Mount Shasta, CA
    As far as using VirtualBox with your Mac and say Windows 7 or 8.1, Microsoft has these free Virtual Machine images available. A complete running OS that you can use for some months before it expires. After expiration you can just reload the VM image to start the clock over again.


  • I tried it on parallels but it is just a kludge which is why I bought the NUC knowing it had no onboard audio,  for feature extraction and sending OSC data to multimedia software running in OSX.

    It's no problem; I am sending MIDI over LAN to an audio sequencer for reward & inhibit sounds. Not quite the data flow scenario I had in mind so will get my teeth into OpenViBE and get that OSC data streams flowing. MIDI might be able to send a relative volume or note value but OSC can send a precise frequency which gives much more scope.

    The only things new to me are the details of EEG work. OSC, DSP, MIDI, sampling etc have been my core work since the nineties and I am very keen to involve what I know about multimedia production.

  • wjcroftwjcroft Mount Shasta, CA
    Doesnt the NUC have HDMI output, which includes audio?


  • Windows says no audio device installed. Makes sense, the motherboard is less than 4 inches square!


  • wjcroftwjcroft Mount Shasta, CA
    Other people are getting audio from the NUCs...


    It's being used in a lot of home theater applications. Perhaps you are just missing a driver or the Windows version you have has a glitch.
  • I could get audio from this one if I connected a USB audio interface, they just don't have onboard audio as standard. It was always the intention to have it feeding multimedia rendering gear over the network, not running the whole show.

    Incidentally, there is no appreciable latency whatsoever. We have been using OSC to send VJ data around huge arenas because it is very responsive on a closed LAN. The overall latency from the output of OpenVibe/Neuromore to hearing/seeing the feedback is around 25ms. The audio engine has a latency of about 7ms, the visuals probably a bit more but sync of the two has never been a problem when done well.
  • wjcroftwjcroft Mount Shasta, CA
    That's good news about the latency.

    It does seem that some NUCs have audio capability (via HDMI connector), perhaps yours is an exception,


  • edited December 2015
    Almost certainly, NUC is a concept not a single product, However mine is not on that list.

    With regard to my initial post I now have no issue with capturing brain activity. It was behaving unpredictably the first say or two, and I'm putting that down to the use of some batteries found in a drawer when the board arrived rather than dodgy leads or poor application of paste.

    With regard to the full band EEG, I'm not quite sure why filtering below 1Hz is the best way to remove the DC offset in this particular. One way to remove a DC offset on an analogue oscillating signal is with a decent coupling capacitor. If it were a hifi application, a value is chosen where the frequency high pass occurs below audible range e.g. 7Hz

    By filtering at 1Hz we lose all of the ILF data which is something I am interested in studying. Has anyone tried a decent coupling cap on the input pins instead?


  • wjcroftwjcroft Mount Shasta, CA
    re: NUC, I still maintain that the NUCs have audio on the HDMI port, since they are a popular choice for home theater applications. The link I gave to communities.intel.com shows that Intel has some issues with drivers and boot parameters, such that the audio device does not show up under Windows unless certain conditions are met. It's frustrating many users, so I can understand your consternation.

    re: capacitors. No. To process full band EEG, you simply have two data flow paths in your element blocks. One raw stream from the channel is conventional EEG and has the high pass filter as the first stream element. The 2nd raw stream from the same channel is your ILF path and can be filtered completely differently. You can choose your conventional path highpass filter any way you like: .5 and .1 hz could be other highpass values used.

  • wjcroftwjcroft Mount Shasta, CA
    More posts on Intel forums regarding audio through the HDMI port.

    https://communities.intel.com/thread/41853?tstart=0

    Lots of NUC users apparently have trouble with getting their HDMI audio to mate with the attached equipment.
  • "re: NUC, I still maintain that the NUCs have audio on the HDMI port"

    I'm sure that some do. My machine tells me there are no audio devices installed, and it is not on the list of those compatible with the Intel HDMI driver so we are probably both right, me specifically and you generally.

    So is that large drift away from zero I was recording in Neuromore a DC offset, or a very low frequency high amplitude signal? Or a mixture of both? I can see how the offset wouldn't be a problem if the ILF channel was separated from the rest of the data. What would be a good frequency for high cutoff on the ILF? 2Hz? Lower?

  • wjcroftwjcroft Mount Shasta, CA
    The term "DC offset" is loosely used. It's not really a steady state DC, but very slowly moving, can go both positive and negative. See the full band EEG thread mentioned above previously for articles explaining it. Typical ILF / ISF protocols look at frequency ranges of say .001 hz to .1 hz. That .001 hz of course corresponds to 1/.001 = 1000 seconds = 16.6 minutes (!) Some folks such as the Othmers even go to .0001 hz.

    No NUCs have an "audio device" on the board. Instead audio is encoded by the HDMI chip and sent out that port, provided a compatible driver and mating equipment is alive at the other end of the HDMI cable.

  • edited December 2015
    Please stop contradicting me and finding fault with my posts. You are not the world authority on BCI or Intel NUC. Some have a headphone jack i.e. onboard audio interface but it doesn't matter becuase i've said a few times now that I don't need it for audio.

Sign In or Register to comment.