Failed project: Opamp Vos tester

[cellularmitosis]

Hello!

I recently completed a small PCB project and found it to be a bit of a failure.  I thought I'd post about it and maybe generate some discussion.

github page, kicad files, gerbers, schematic, etc: https://github.com/pepaslabs/opamp-vos

The idea was to make a simple board into which I could plug in various opamps (it supports three different footprints), and use some DMM probes to measure the input voltage offset error (Vos).

Unfortunately, I'm getting strange readings.  For example, I measured about 7mV offset on an LM358 (using a 6.5-digit Keithley 196), but if I flip my probes, I don't get -7MV, I get about -2mV.  Fail!  The measurements should be the same!

I have a guess at what the problem is: I think I'm also measuring noise, and that noise is swamping the Vos signal.  I switched the DMM to AC and measured about 4.5mV, so this seems plausible.

I noticed that several circuits I found online for measuring Vos all amplify the Vos by quite a bit (i.e. 1000x).  At first I had assumed this was just to allow taking a measurement from cheaper equipment (e.g. a 3.5-digit DMM), but now I think it might also have to do with amplifying the signal to overcome the measurement noise.

Thoughts?

[cellularmitosis]

(attaching schematic)

[danadak]

This is a long shot but is 196 out of cal. Do the probe flipping routine
on another dc source, like a battery.

Also thinking possible noise pickup on NI input. Note your circuit aggravates
Vos due to unbalanced source Z and Ibias presented by voltage divider at input.


Regards, Dana.

[danadak]

This might be useful -


http://www.analog.com/media/en/training-seminars/tutorials/MT-037.pdf?doc=CN0326.pdf


Regards, Dana.

[T3sl4co1l]

Electronic paranoia step 0:
Put bypass caps anywhere you can afford to.

(This is just an equivalent form of the more general: "Use only the bandwidth you need.")

Your 1.6V bias level doesn't have any bandwidth (it's "DC"), so you can afford to put even a huge amount on there.

0.1uF would be fine.  The main offending sources will be RF that can get into the circuit, i.e., wavelengths proportionate to the size of the circuit, or any wires hanging off it.

Same goes for the voltmeter sense pins!

If you're measuring the amp output directly, obviously don't stick a cap there, but add a series resistor (typically 100 ohms) and then the cap.

Tim

[cellularmitosis]

Dana, thank you so much for that ADI tutorial on Vos!  That looks super useful!

I took a shot at measuring how much "lead flipping" error is coming from my (out of cal) 196:

• I used my other DMM as a voltage source by performing a 2-wire Ohm measurement (against a 4R7 resistor)
• covered the resistor, etc with a dish towel
• flip the leads one way, wait a minute to settle: ~4.811mV
• flip the leads the other way, wait a minute to settle: ~4.795mV

Interestingly, a polarity flip does cause the DMM to disagree!  However, the difference is under 20uV.

I flipped over to AC and it was picking up ~70uV of noise, so the offset could be buried in there.

[cellularmitosis]

Apparently I can also stand on my head while taking pictures! 

[cellularmitosis]

Tim, thanks for the ideas!

I made a few modifications:

I changed the divider to 1:1 (2.5V at the input)

I tried bodging in 0.1uF caps at the opamp output. (see attached)

Here's the equivalent schematic of just the part I am testing: (see attached)

Oh wait, what's that they say about opamps?  They can't drive a capacitive load?  Oh, right, derp!  (see attached)

Should I try moving the caps after the 1k resistor?

I suppose ideally I should try cutting traces and put a resistor inbetween the opamp output and input...

[SeanB]

Twist the test leads together, and put a 100R to 1k resistor in series with the cap on the output. Also put the whole thing in a metal cake box ( or any other large metal box, with a ground braid connecting the lid to the box, soldered on both sides, and ground the box to the common lead), with 100uF of capacitor on the supply, along with winding both sets of leads ( power and test) through a ferrite toroid.

I would also use some gain, around 2x at a minimum, as many opamps are not unity gain stable, and as you are only looking at DC gain put a film capacitor ( 1n at most) on the offset pins to limit bandwidth there as well. Gain as well gives you a better offset figure to measure. Reference gain resistors against your split rail, and decouple with a film capacitor there as well, and use some series resistance to the non inverting input ( roughly the same as the gain side, around 10k) to approximate closer the manufacturers test rig.

you probably also would want to make the supply variable, using a lm317 or such, so you can test at various supply voltages.

[Cerebus]

I took a shot at measuring how much "lead flipping" error is coming from my (out of cal) 196:

• I used my other DMM as a voltage source by performing a 2-wire Ohm measurement (against a 4R7 resistor)
• covered the resistor, etc with a dish towel
• flip the leads one way, wait a minute to settle: ~4.811mV
• flip the leads the other way, wait a minute to settle: ~4.795mV

Interestingly, a polarity flip does cause the DMM to disagree!  However, the difference is under 20uV.

Dissimilar metals can, and will, act as thermocouples, generating a few uV from the temperature differences between parts that have been recently handled, or are self heating (my bench multimeter front panel is currently 3 C above ambient).

Normal metrology practice for low level measurements that will be significantly affected by this is to measure, then flip the measurement setup (as you have), flip the sign on that 2nd measurement, average the two measurements and Bob's yer uncle, the thermocouple effects disappear as their polarity doesn't change when you flip the measurement.

[danadak]

An ADI ap note on various basic errors -


http://www.analog.com/media/en/technical-documentation/application-notes/294542582256114777959693992461771205AN280.pdf


Regards, Dana.