Ganglion Input-Referred Noise?

GregoryGGregoryG Paris, France
edited October 2017 in Ganglion
Hi,

I am interested in using Ganglion or Cyton board to record Auditory Brainstem Response to click stimulus. To determine the feasibility, I need to know the Input-Referred Noise measured when the input is shorted together. For the Cyton board, all is in the ADS1299 datasheet on Table 1 to Table 5.

For the Ganglion board, I found 2 measured values on the forum.

The first one give a value of 1uVpp Input-Referred Noise at Fs=256Hz ("10 Minutes Noise Test Ganglion.txt")

https://github.com/OpenBCI/Ganglion-Prototype-Data

The second one give a value of 10uVpp Input-Referred Noise at Fs=256Hz

http://openbci.com/community/ganglion-review-input-noise-and-p300/

In both measurements, unfortunately the level of PGA gain is not specified... and it's obvious that gain will influence Noise value. Does someone know the value of Input-Referred Noise regarding the sampling frequency and PGA gain ?

Thanks,

Comments

  • wjcroftwjcroft Mount Shasta, CA
    Mentioning Joel @biomurph and @openbci.

  • wjcroftwjcroft Mount Shasta, CA
    Gregory, hi.

    I just re-read Alex's page at,


    I do not see where he mentions the 10uV noise figure? Can you clarify? His measurement is 1.5 uV as stated in the Input Noise section directly underneath the graphs.

    Perhaps you were confused by his statement later in the page, "The signal looks okay, with some alpha burst at -8 and -2 second. I have around 10 uV of standard deviation, which is usually the case for EEG." This is not a noise measurement.

    I believe the gain is set at default values. All the figures are in microvolts. Mentioning @openbci.

    Regards,

    William

  • GregoryGGregoryG Paris, France
    edited October 2017
    Hi William,
    Alex mentions 1.5uVRMS for the Input-Referred Noise at Fs=256Hz, but if we look at his distribution graph (the left one), we can see peak to peak around 10uVpp.
    For me, either VRMS or VPP is ok, I can deduce approximately the another one. 
    I just need to be sure about the PGA gain but you are probably right, it must be with the default gain, so GAIN_1 right ?

    Just to compare, the ads1299 of the Cyton have an Input-Referred Noise of 1.08uVRMS at Fs=250Hz for PGA gain of 1.
    Regards,



  • The gain is hardset on the Ganglion to 51 i think, @biomurph understand this more!
  • Yes, on Ganglion there's a hard-set mid-band gain of 51, in the instrumentation amplifier stage.
    Then the signal goes into the AFE, the first stage of which is a PGA for each channel, SPI-command controllable for gains of 1, 2, 4, 8, 16, or 32.

  • GregoryGGregoryG Paris, France
    Indeed after discussing with Alex, the measurement of 1.5uVRMS for the Input-Referred Noise at Fs = 256Hz was done with a gain of 1 on the PGA of the MCP3912.
    He also told me that the gain of the MCP has no influence on the noise, because there is a stage of pre-amplification (gain 51) before the ADC, realized via the AD8237.
    He is absolutely right cause this gain of 51 already gives peak to peak amplitude well above the ADC INL.

    To return to my original question, I need to know the Input-Referred Noise measured when the input is shorted together, but I found 2 measured values on the forum.
    The first one give a value of 1uVpp Input-Referred Noise at Fs=256Hz ("10 Minutes Noise Test Ganglion.txt")

    The second one give a value of 1.5uVRMS Input-Referred Noise at Fs=256Hz

    Can @biomurph could help us to understand these differences ?
  • Hi GregoryC,

    I think the mentions for @biomurph or @openbci or @somebody are good.

    ........
    Regarding the 10-second dataset at https://github.com/OpenBCI/Ganglion-Prototype-DataI'm thinking that without more contextual info, those data are almost worthless, maybe even too-easily-misleading. 

    Seems I or someone in my situation would at least need to know the nature (at least a schematic) of the "Ganglion Prototype" that was used at that time >2 years ago.  I'm not too involved now and sure wasn't back then, and for all I know the MCP3912 wasn't yet used, or there was something different from the AD8237 instrumentation amp, or a different IA gain, different V_REF/etc ckts, different highpass-pole (I think we know that has changed), different D_G (called DRL in the README.md) circuit, etc..  In fact it's worth noting that the readme and the file header say "Sample Rate = 256.0 Hz" ..... which isn't a usual-normal sample rate for the production-released Ganglion.

    On the plus side, according to the README.md file it does make some reference to the hardware-interconnections:  "All inputs tied to DRL".  So if that's correct, then there's no external connection to circuit-ground (GNDA), and there's no simulated-electrode-impedance (which may be less realistic for a user, but inclusion of electrode impedances wouldn't be likely to affect noise magnitude anyway).  Other internal/external connection methods are useful to do, especially during development, but mostly what's useful for a user is external-only connections, similar to connection to a patient.

    .......
    If those reported noise-data were questionably-high, then it would also be important to understand some things such as the electromagnetic environment of the measurement.  But to me they seem questionably low.  

    First of all, sort of a side note ....
    In that 10-second file, for channels 1, 2, 3, 4 I'm seeing 1.34, 1.32, 1.41, and 1.40 uVpp respectively.  So (GregoryC) am I right that you were simply approximating/rounding when you said "1uVpp", and we shouldn't get hung up on that number vs something like 1.4uVpp?

    Further on .....
    Looking at things more analytically, you might follow this depending on what experience you have with noise- and frequently-analysis.  Note that for Ganglion, the RC in front of the A/D + inputs rolls off at about 160Hz, plus there's rolloff or aliasing that may occur due to sample method/rate, but let's say 160Hz.  Looking at the AD8237 instrumentation amp, the broadband input-referred noise (which would be the differential noise between pins 2 and 3) is typically 68nV/rtHz; and looking at the spec's Figures 30&32, you might make a rough estimate that the 'averaged-like' value traversing up to 160Hz might be about 80nV/rtHz.  Those are rms voltages, so the resulting noise would be 80n * sqrt(160) = 1.0 uVrms.  

    That's not the only noise source, but it's plenty to call into question the approx 1.4uVpp values from the 10-second file taken on the Ganglion prototype.  I think two other noise sources worth considering are:
    -- The AD8237 amp's "1/f" noise, shown graphically in the spec Figure 31
    -- U6 opamp noise, forming the V_REF signal on its pin 7.  Speaking in terms of the 1st channel, I think it's not a problem for the MCP3912 negative-inputs such as CH0-, because you can divide by 51.  But I don't think you get to divide by 51 for its effect on the FB circuit, nor for its effect on the AD8237 REF_1 circuit.

    Do keep in mind that when you combine (uncorrelated) noise sources together, you don't want to add them algebraically; you should add them in RMS-fashion, that is, Vtotal = sqrt( V1^2 + V2^2 ).

    ........
    Putting it all together without better calculations/analysis, the 1.5uVrms level might sound a bit high to me, but not far off.  Unfortunately I've never done Ganglion noise testing and don't currently have access to a system to do my own, I've just thought Alex's posting sounded reasonable.

    .......
    Two final caveats, though:
    -- It seems to me that by Ganglion design, the A/D is being operated in an unspecified region, so among other things we can't trust its noise specs (nor its INL by the way, but INL is in some ways separate/in-addition-to noise effects).  I posted about the unspecified-region before, no responses yet, don't know whether I was clear enough or what, but maybe you can see it at http://openbci.com/forum/index.php?p=/discussion/1406/ganglion-hardware-design-flaw-misunderstanding.
    -- In Alex's blog at http://alexandre.barachant.org/blog/2017/03/02/Hands-on-OpenBCI-Ganglion.html he says the noise testing that he did can be "achieved by connecting all inputs (and reference) to the ground".  With the word "ground", I'm unclear whether he means to Ganglion's GNDA, or to something else.  If he means to GNDA, that's not necessarily bad, but it's a somewhat different noise test than what you might want and different than the 10-second-test supposedly did, and without looking further I'm unclear whether that might alter noise.

    Good luck!  Hope you hear from those who are knowledgeable about prior testing & all.

         -- Bruce P.

Sign In or Register to comment.