What should this op-amp Vos test jig do in LTSpice??

[MathWizard]

I made this full test-jig on a PCB that can measure Vos among other things. And I did the full calculations, (but not with input bias currents yet) and seemed to get the right answer, at least if the op-amp's gain goes to infinity or the frequency goes to zero, I get Vo2=-Beta*Vos1=-1000*Vos1. But I don't trust what LTSpice is doing, but the real circuit seems to work and give typical Vos values for the few op-amps I tried so far. I haven't tried any other test modes yet.

IDK that many op-amps in stock LTSpice, but if the circuit sim really isn't working, I don't get why. I've set the initial charge on the cap to zero, tried short and long sim's. For instance with a stock LT1013, my own seemed to measure as 40uV, with Vo2~40mV, which is spot on typical. But in LTS, Vo2 is 1uV, so ignoring bias currents, they are saying the LT1013 has a Vos of 1nV. Of all the op-amps I tried in LTS so far in this circuit, the highest Vout2 I've seen was ~6mV, so |Vos|~6uV.

So what's gone wrong ??

[MathWizard]

Now the circuit IRL, works ok sometimes, but other times, it has oscillations, and the open-loop gain of the DUT op-amp, makes it's output jump a lot. The tests for input bias currents +IB, -IB can work ok sometimes, I've gotten legit figures that match datasheets.

I'm using some brand-name sample book SMD caps, for the 10uF cap. Looking at the eqn's I got, and the sim, I see the circuit acts like a LC-low pass filter. And I was having some 120mVpp 12Hz osc. when in Vos mode. So I added a 4.7uF cap in parallel with the 99.9k Rf. And it seemed to be nice and stable, for about 1minute. And then the output goes walk about, not any nice 12Hz sine-wave tho.

I used 2x100nF caps on both DUT and AUX op-amps, and I have 10uF on both rails. And both rails are behind a power switch.  And I have very small size switches (expect voltage levels) all in place. I pretty carefully choose all the resistors, but most of them are just cheapo 5%, since I have a huge bag of 100k's and 100's, and could use a single resistor.

The layout is fairly neat and besides stray capacitance, I'm guessing loops don't matter at all for the 12Hz from the 110k and 10uF cap.

But the Rf resistor is over an inch, or 26mm, away from the op-amp socket. I've scrapped and cleaned all the flux off the proto-board. There's a chance there's RA flux under the 4 SMD caps I used, and I know this flux has issues with leaving a conductive sludge.



But I find this typical of op-amps, where I'm not sure if it's my probes sometimes, but the DMM see's it too. But the output just goes off all over the place sometimes. And they don't warn about this in the pdf, or use as detailed of eqn's as I did. But IDK how else to fix this circuit.

But I did leave the other 1/2 of the PCB empty, to add some Texas Inst. versions of these same tests. It will be nice to have another version to compare with, right on the same PCB.

[Kleinstein]

Many, if not all OP-amp models have no offset voltage included. If one wants an offset one would have to add it.

If the typical specs for the offset are 1 mV one can not expect an offset of 1 mV as a typical value. Instead this means that most of the OP-amps (e.g. 90% - not sure about the exact number - this may even vary with manufacturers and can change over time) have an offset between - 1 mV and + 1 mV.  The typical offset specs are still a kind of upper limit and there tend to be positive or negative values and also values close to zero. For a few types they show distributions of parameters that they actually measured for some test batches. These are usually early units and later performance can change a little.

It gets a bit more complicated with input bias: here it depends on whether there is compensation or not. So there could well be an average bias current that is give. With compensation there are still often both signs  possible and the number should be more like a typical upper limit for the absolute value.

For the test jig circuit it would help to have some resistor (e.g. 10 K range) in series to the capacitor. This should help with stability and give less ringing.   

[MathWizard]

Ok I was wondering if it was some ESR thing, but I only tried adding 0.1 ohms. So try adding more damping. I've already paired up a pile of 100k's and 100's, to make this TI test jig, both the single and dual op-amp version.

Ok thanks Kleinstein, between cleaning under those caps, and then using waxy flux, and using a 50k pot that I never adjusted and reads in circuit as ~3k, it seems to be working pretty good, w/ a DMM.

The only funny thing I've seen lately was the Vos of a LT1013 jump from ~40uV to 20uV, after I throw a switch to measure one of the bias currents. It happened twice. Or maybe just changing the range on my DMM from 20V to 2V did it, but it's still on 10G ohms input setting.


Oh no nevermind, looking with a scope, there's now a rail to rail 12Vpp, 16kHz triangle wave, with the top clipped off.

[MathWizard]

That was working last night, I just probed the side I didn't put the caps back on. Now it's pretty stable, but the Vos measurement still moves more than the IB+/- settings. I put it in a GNDed, tin can, and I can't really see any difference w/ a DMM.

I still don't trust the LTSpice transient results that much, or really I'm not having it do what I want/expect, like with a universal op-amp rail and setting the Vos. I don't think it's the bias currents, they are much less than in real life.


But I see I must have some fake TL072's, bought years ago on Ebay. Their Vos is good, but the input bias current is 11nA (taking their absolute value 1st), and their Ios is 22nA. For JFET's the datasheet gives 5-100pA Ibc. But trying to measure the Vos, the output spends a good while at 0.36xxV, then it will spend some time at around 0.42xxV, then back. I don't read any amount of AC on the DMM, <1mV most of the time. So IDK how much more of that can be fixed.

[magic]

The cheapest of opamps sold on eBay/AliExpress/Amazon/etc are Chinese LM358 or RC4558 clones, regardless of what they "identify" as. The TL07x are usually LM358, while RC4558 is more common in typical audio parts, maybe because customers complained about crossover distortion and noise of the 358

[MathWizard]

I checked some SMD TL082's that I thought were from a legit source, but they also seem to have ~18nA Ibc. This test jig will is not near stable enough to measure pico-amps tho anyways. I'll try finding or making a much bigger unpolarized feedback cap on the 99.9k resistor.

But overall it's working fine now, and is giving usable results. Now I'm using Vos and the input bias currents in calculations, so I need to make something and see if it works as it should.