That is indeed a fair amount of drift, the rising output. So my first question is how are you measuring this voltage? What device are you using and just where are the measurement points located? I ask this because I have seen folks chase errors and drift that were not there when the measurement system was disconnected. If the rising voltage actually appears in the A/D count then it matters a lot more. Another important question is: Does the indicated voltage increase when the signal connections to the A/D are replaced by a 1K ohm resistor? In the flowmeter circuit is a block with six leads, and no information at all. Is that a passive device with no power conected to it? Or does it contain Some active device that is powered? If it is only an RC filter then I would point at the opto isolator, since it is DC coupled to the A/D through the filter circuit. So there are some questions to answer and checks to do. Those answers will be useful to all readers here.

Sorry it looked like the other questions were put in after I quoted and responded to the original post.

I put a 10K resistor in parallel with the 2x analog inputs (between the ADC input and ground), and at first I thought it solved the problem... but sure enough it started to come back. Sometimes it comes back on its own, and sometimes me tapping the wire to activate the flow meter makes the problem come back instantly.

You mentioned a block with six leads, are you referring to part U$18? This is labeled as a LTC6993IS6-3 which is a monostable multivibrator. Its also labeled as being powered with 3.3V. Its used to normalize the pulses - any pulse that comes, is converted to a fixed width pulse, this way I can get consistent measurements.

And also like I mentioned previously, I can take a voltage reading on the left side of R76 (between R76 and the monostable multivibrator U$18) and I get 0 volts. I can clearly see the drifted voltage on the right side of R76 (between R76 and the ADC), so it seems the voltage is coming out of the ADC doesn't it?

 

your thinking/setup/biasing is "technically" correct - you can reference both inputs to the same reference point no matter would the reference point be static or varying in time . . . exept the common mode sensitivity varies ... a bit with your reference point changing + the Op Amp has an offset error . . . those two can lead to nonlinear output

other than that -- some Op Amps may have 10s of minutes . . . yes it's minutes of the "warm up time" -- so you should read the docs if the one you are using is such

I can clearly see that inbetween U$18 and R76 is 0 volts, which is what it should be when the flow meter is not running. This part is correct, and there is no Op Amp here. On the right side of R76 (between R76 and the ADC) I can see the drifted voltage e.g. 20mV to 80mV. This seems unrelated to any Op Amp error.

 

I can clearly see that inbetween U$18 and R76 is 0 volts, which is what it should be when the flow meter is not running. This part is correct, and there is no Op Amp here. On the right side of R76 (between R76 and the ADC) I can see the drifted voltage e.g. 20mV to 80mV. This seems unrelated to any Op Amp error.

I was unable to read any of the text on the drawing, hence my question.And since the voltage does appear as you describe, it may indeed be that the error voltage is caused by either input bias current or input offset voltage on the A/D converter. Other than suggesting a unity-gain buffer amp I have no other ideas to offer. But there are some smart folks who may already have solved the problem, possibly they can share a fix.

 

unable

now enabled

U$18

? flow meter ? what offset ? you use an opto with nonlinear CTR and temperature dependency to isolate/transfer the analog signal ??? . . . i must re-read the first post
. . . so it's the adc offset of the micro . . . how about calibrating on the run from GND and https://www.ti.com/lit/ml/snvd001a/snvd001a.pdf?ts=1588623293875

the possible scenario for such might be adding a negative supply (perhaps a charge pump converter) and positive above Vdd then you can add intercept diodes and amp to force the DAC input pin to the GND or pull it up to V.REF ← assuming it's never reached by your inputs or to "ideal Vdd" . . . requires some "absolute" (read very low error) referencing . . . you likely not interested adding to your design .... anyway it's doable

 

Quote: "I can clearly see that inbetween U$18 and R76 is 0 volts, which is what it should be when the flow meter is not running. This part is correct, and there is no Op Amp here. On the right side of R76 (between R76 and the ADC) I can see the drifted voltage e.g. 20mV to 80mV. This seems unrelated to any Op Amp error.
OK, given that we have no error on one side of the resistor and error on the other side, how about a jumper across the resistor?What I thyink I see is that the lower source impedance reduces the error voltage to a point below your measuring ability. THAT might be low enough. It will be a super cheap fix if it works. And not cost much to try.

 

I can clearly see that inbetween U$18 and R76 is 0 volts, which is what it should be when the flow meter is not running. This part is correct, and there is no Op Amp here. On the right side of R76 (between R76 and the ADC) I can see the drifted voltage e.g. 20mV to 80mV. This seems unrelated to any Op Amp error.

Sounds to me like maybe the ground reference for the ADC is drifting and/or being driven by something.

Can we see your PCB layout? Is there significant current from some other device to ground that could be pushing local ground at the ADC up?

Check for loose or intermittent connections or maybe cold solder joints at the ADC ground pin, and also at any possible failure points between the ADC ground and power supply. (I've seen cold solder joints that worked well enough to go undetected for days, but also yielded inconsistent results over time. Don't fall into the trap of thinking joints are either obviously good or bad - there are plenty of spotty, not so obvious bad ones out there.) Bad vias are another possibility, assuming it's on a multi layer board.

 

so it seems the voltage is coming out of the ADC doesn't it?

(( i don't want to start reading another electrical spec.-s for the MCU I/O -- that usually are lacking (full) detail ))
if you are right you just answered your own question