• State
  • Locked Locked
  • Replies 7 replies
  • Subscribers 32 subscribers
  • Views 2441 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

how to reduce ADC noise

george dorian
Intellectual 940 points


Hello all,

I'm not huge in A2Ds and have some Qs about reducing noise.

OK, Signal to Quantization Noise Ratio is: SQNR = 20log10(2^n)

Where n is the number of A2D bits.  So with a 12 bit A2D you will have about 72dB signal above quantization error (one least sig bit).

Now, what happens to the SQNR if you do some averaging?  I think you get 3dB improvement per 8 over samples.  Is that right?

And what happens if you right shift out the least sig bit (or two)?  Do you just get a 11 bit A2D (with about 66dB SQNR) or do you actually get better SQNR.  

How can you write code to make a firmware anti-aliasing filter? 

Is it true that you still need an analog anti-aliasing filter even if you do it by code?

thanks george

  • 0
    Guru 47900 points

    May I - via our (long) experience w/such MCU based ADCs - steer you in a (bit) different direction?

    If so - while we appreciate the precision of your "SQNR" focus - the "real world" of MCU ADCs presents a harsh reality.  The twelve bits of resolution exist (primarily) in fantasy-land - not w/in this class of mixed signal device.   (this is no knock on this vendor - our firm employs ARM MCUs from at least 4 vendors - all behave/perform {in general} similarly.)

    Our experiments - contrasting our best efforts w/such class MCUs - reveals the 3 lsb to (most always) be suspect.

    I'd suggest that your best/fastest measurements may result from:

    a) very careful pcb design - following all of the various analog signal routing & separation from clocks & higher current signal routings.

    b) employ a separate supply for the analog Vdd.   If that's not possible - we find improvements via the insertion of a ferrite bead & both high & low value capacitors as close to the  Analog Vdd pin as possible.

    c) make your measurements - if possible - when the highest current events are unlikely or at minimum.

    d) old standby of shielded cable often improves (especially) low level signal measurement.

    e) comply with each maker's design recommendations - especially matching ADC input impedance.

    Note that a single capacitor is charged with much of the ADC's measurement challenge - imbalance via too fast or too great signal variation reduces conversion accuracy.

    These devices do provide hardware "averaging" - you trade speed of conversion for the "masking" of reasonable transient signal levels.

    Your presentation & concerns - of course are valid - but my sense is that these (primarily) HW basics (as listed) deserve your first consideration.   Even the best SW cannot save a flawed analog design...

    To your quest for precision - might your review of the "pro" ADCs offered by this vendor (and others) provide additional insights?  Look carefully at the design of their "16b and up" ADC device Eval boards - much to be learned - and modeled there.   Your world of SQNR - and optimizing tactics - has far more potential for measurable success in that (pro ADC) sandbox...

  • 0 in reply to cb1
    Guru 27665 points
    Agree with cb1. A couple of additional notes
    - TI (and others) provide seminars on interfacing to A/Ds. They are worth attending. At least TI's was definitely worth attending, they did address the analog niceties that get overlooked in the data sheet presentations. The information is in the data sheet but not obvious if you are not steeped in A/D practice.
    - If you want absolute accuracy you are going to need to add a reference to your circuit. Some things like hall sensors and potentiometers are fine using the power supply for a reference, other sources are not.
    - You need a charge reservoir to deal with the time scale of the mux. With a multiplexer the sampling time is not the inverse of the channel frequency, it is much faster.
    - You need antialiasing to keep frequency components higher than supported by your sampling rate out of the signal. You can filter lower.

    Finally, this may all be moot. Check your signal source. Signal sources more accurate than 10 bits are rare. Many are less than 8 bits.

    Robert
  • 0 in reply to Robert Adsett
    Guru 47900 points

    @Robert,

    Good stuff - poster should be pleased.

    I'd note that even with a good, rock-solid reference - we still have jitter in those 3 lsb. (surely improved - but those bits are (almost) meaningless...)

    As for, "signal source inaccuracy" - might the "difference" in two signal levels escape that particular concern?   And prove of great use/value...
     
    Lastly - applaud your mention of A-D seminars.    And - surprisingly - attending seminars provided by multiple ARM vendors - often provides key insights into any/all weaknesses of your/my chosen device!    (competitors surely "know this" and are "not shy" about "parading the weaknesses" of other vendor's devices!)

    As such "negative" info is unlikely to rise from any vendor source - attendance at your chosen vendor's AND competitive vendors' seminars - may prove of great value!   (surely has for us - leapfrogging is constant - "single source" has long been branded as less than, "best/brightest!")

  • 0 in reply to cb1
    Guru 27665 points
    @cb1
    Good point about parading weaknesses, although they do not usually name names. The vendors with a long history of solid analog development are usually the best bet for informative seminars of this type. They usually get into signal conditioning a bit as well.

    The difference between two signals might be more accurately measured, that's the basic principle of a resistor bridge circuit after all, but it wouldn't always work.

    I've recently obtained a good solid calibration source but haven't had the opportunity to verify performance. I have observed the result does appear to be stable to 11 bits or so but that is after heavy filtering.

    Robert
  • 0 in reply to Robert Adsett
    Guru 47900 points

    @Robert,

    Your noted, "appearance of stability at/around 11 bits" (for a 12 bit spec'ed device) exceeds our (long) and best findings! Bravo.

    Might you share some details regarding your, "calibration source."   We usually employ vendor's Voltage References - find those (most always) "Good for Gov't Work!)

  • 0 in reply to cb1
    Guru 27665 points
    Don't be too impressed by that stability, it's largely a function of heavy filtering after the conversion.

    This www.voltagestandard.com is where we purchased the calibration source (specifically the Pentaref). Inexpensive and it is more accurate than I expect the A/Ds to be. I haven't really put it through its paces yet but so far it appears to work fine.

    Robert
  • 0 in reply to Robert Adsett
    Guru 47900 points
    Merci, monsieur.