I'm building a circuit that has a 2.5V Voltage reference for use on the internal ADC of a STM32F7 in 12 bit mode. As each bit corresponds to 610uV I'd like to have as little noise in the signal as possible. The voltage reference is also adding 1.25V DC offset to the incoming signal, as the signal is partially a negative voltage, so I have a 1kohm by 1kohm voltage divider on the output.

The voltage reference is TIs REF5025 datasheet here: http://www.ti.com/lit/ds/symlink/ref5025.pdf

I'm using the voltage reference in its simplest configuration, 10uF Cbypass and 1uF CL; from the datasheet: enter image description here

My issue is that I can't measure how much noise is genuinely on the output of the voltage reference. It is outputting 2.5V as expected but I also wanted to see how "clean" the DC out is compared to my bench power supply that is currently being used for these tests (the final circuit is using an LM7805 as the 5V supply)

Channel 1 is the power supply (yellow) and channel 2 is the voltage ref (blue). as you can see, the power supply actually has LESS noise than the voltage reference (approx. 10mV compared to 20mV).

enter image description here

The power supply is cleaner because my ground connection is as short as possible so I'm picking up less measurement errors/ background noise I'm guessing.

enter image description here

So I believe that this is just measurement error and the noise is from my probe acting as an antenna and picking up background noise.

If I wanted to measure the genuine noise coming from the voltage reference how can I do this?

The datasheet states 3.5uVpp/V, how can I ensure that I am getting this accuracy (or more realistically, < 1mV noise?

How can I ensure that I am at least getting a reference accurate enough for an ADC resolution of 610uV per bit?

  • \$\begingroup\$ Make a moderate gain differential amplifier to amplify the noise by (say) ten or 100. \$\endgroup\$
    – Andy aka
    Apr 1, 2020 at 15:52
  • \$\begingroup\$ Do you have access to a spectrum analyser? AC-couple the REF signal to its input (with coax). You might also try analyzing a X1 scope probe too, with its probe tip shorted to its ground alligator clip...so you can see any local magnetic-field noise sources. \$\endgroup\$
    – glen_geek
    Apr 1, 2020 at 16:18
  • \$\begingroup\$ Could you explain a bit more? Are you suggesting amplify the output of the voltage reference to reduce the ratio of ground noise to genuine signal. Then any genuine noise on voltage reference would also be amplified. Am I in the right ballpark? Won't adding an op amp also introduce it's own noise component? \$\endgroup\$
    – ChrisD91
    Apr 1, 2020 at 16:20
  • \$\begingroup\$ Unfortunately no spectrum analyser access, just a scope and multimeter as I'm working from home (lucky enough to have a hobbyists lab) \$\endgroup\$
    – ChrisD91
    Apr 1, 2020 at 16:22

2 Answers 2


For good noise measurements, there are a handful of things you can do to clean up what you see on the scope:

  1. Use a 1:1 probe instead of a 10:1 probe.
  2. Use the lowest voltage range possible
  3. Reduce the bandwidth setting (if the noise is in the lower frequency, 20Mhz typical)

More info on the above here and here.

Also, reduce or eliminate the ground clip lead.

enter image description here

  • 1
    \$\begingroup\$ Making the probe x1 and wiring up a ground point in the probe like in the image, along with removing any potential antennas helped massively. My power supply does indeed seem to have about 10mVpp noise but the Vref I can now measure at about 1mVpp. I'm assuming that if my scope has a lower resolution than 1mV or if I amplified the signal I could get more accurate result. Thanks very much, I can now prove that the Vref is a much more accurate reference than the power rails. \$\endgroup\$
    – ChrisD91
    Apr 1, 2020 at 19:19

Run a coax cable from Reference IC to your scope.

Examine the scope display, with and without 47uF on the REF output.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.