EEVblog #476 - Opamp Offset Voltage Measurement

[EEVblog]

Dave investigates an issue with the input offset voltage on the Analog Devices AD8628 zero-drift chopper opamp.
The answer will be presented in the next video.

[Slothie]

Since the +rail bypass cap seemed so critical I was surprised you didn't try a bigger one to see if that had any effect.

Other than that I cant say - to me the 741 is a mysterious new-fangled device!!

[mswhin63]

Still good enough to be a fundamental Friday  .

What was you output load?

[EEVblog]

Still good enough to be a fundamental Friday  .
What was you output load?

Just the multimeter. Thought I mentioned that? maybe not.
Shooting the real FF now...

[Alexei.Polkhanov]

In an OP-AMP Application Handbook published by AD they a circuit for measuring offset voltage (attached). Perhaps bias current is the reason for inconsistency here because 1K resistor is only present at inverting input.
 

[c4757p]

In an OP-AMP Application Handbook published by AD they a circuit for measuring offset voltage (attached). Perhaps bias current is the reason for inconsistency here because 1K resistor is only present at inverting input.

Well, at the maximum 100pA bias current, and about 1k on the inverting input, you get an induced bias of 100nV. I don't think that's it.

[Rufus]

I suspect the issue will have something to do with common mode voltage as well as supply voltage.

Would be interesting to put your reference on the wiper of a pot across the rails (with some decoupling) and see what happens.

I imagine it is something to do the amount or imbalance of charge injection on all the CMOS switches varying with rail and/or rail to signal voltage. I could see the high frequency impedance of the supply rails having an effect on that. Might be interesting to see if a couple of ferrite beads on the rails (to increase the impedance) makes a difference - might even help.

Imbalance of impedance feeding the inputs might make a difference, sticking a 1k in series with the other input might be worth a try.

[Alexei.Polkhanov]

Analog Devices have "Multisim Component Evaluator Analog Devices Edition" free to downloat on their website, which presumably have accurate SPICE models for AD devices. I used it once before and model was pretty accurate.
Results for single rail supply seem to match what David has on video. With 5V supply it shows 400pV output, but if I lower supply to 3V it will go up to 35mV. When I tried dual, positive negative supply configuration simulation shows something completely unreal - output goes to +2.5V. It can be because I did something stupid in my simulation or it is really trying to tell me that this OPAMP won't work with dual supply.

[SeanB]

Ground needs to move to other side of R1, otherwise you are measuring the full gain of the opamp amplifying the voltage generated across the 1k resistor by the input bias current.

[toli]

In an OP-AMP Application Handbook published by AD they a circuit for measuring offset voltage (attached). Perhaps bias current is the reason for inconsistency here because 1K resistor is only present at inverting input.

Well, at the maximum 100pA bias current, and about 1k on the inverting input, you get an induced bias of 100nV. I don't think that's it.

This current actually flows through the 100K and not the 1K (it can't go through the 1K since the voltage across this resistor is the offset voltage, and doesn't depend on the bias/offset current). The reason we put a (1K||100K) resistor at the non-inverting input to get rid of the offset caused by Ibias is that this voltage will than be amplified by 101.
So the gain for the bias current from the inverting input will be -100K, and from the non-inverting will be 101*(100K||1K) which yields 100K and the two gain cancel the effect of Ibias

BTW, as Rufus has suggested, there's a difference in the input CM voltage when you switch between dual and single rail. Adding a simple voltage divider between the two supply lines will keep it in the middle of the supply range and will remove this effect from the measurements.

[EEVblog]

Next video has already been shot, but I'm going to be an evil bastard and hold it over until next Friday 

[mimmus78]

How much donations you want to publish before next Friday?

:-)

[EEVblog]

How much donations you want to publish before next Friday?

A day off next Friday sounds pretty priceless to me 

[Bored@Work]

Ok, you are applying a wrong definition of input offset voltage. Input offset voltage is the voltage you need to apply to the two op-amp inputs to get a zero output.

[IanB]

By the end of the video I was definitely looking to see a 1 k resistor on the +ve input between pin 3 and ground. What difference does it make to the results if you insert such a resistor?

[TheWelly888]

Dave - you got me looking at the chopper amp video again and in that, you made the point that there needs to be a load across the output of the chopper amp. The only load I can see across the output is the multimeter which I assume is 10M ohms. Maybe try a 10k or 1k load on the output and see if there is any effect?

[ddavidebor]

Baaad foundamental friday.

No, solution, no teory, no whiteboard.

I suggest you to publish the new one sooner, this was a thumbs down for me.

And please do a mailbag, you've alredy 2 sagan of mail to open.

[kkp]

..And it's not thermocouple effect either. If so, it wouldn't change that quickly with supply voltage.
..And the CMRR is typically better than that.

I have seen this before; It can happen when you connect an excessive capacitive load and/or antenna (meter lead) directly to the output of the amplifier.
Squinting at the video I can't make out the print on R9 (the output series resistor Dave likely put in the ucurrent for the same reason), it is populated, but the print looks a bit short to be '101', and it's not in the DaveCad*.
A bit of capacitance across the feedback resistor would probably be an idea, too. There should be 500mV p-p noise coming out of it without one.

Come back next week to see if our AD-caped hero defeats the evil Doctor Offset.

*Generating netlists in this is a pain.

[firewalker]

Could you try lower values to the feedback network?

Alexander.

[EEVblog]

There should be 500mV p-p noise coming out of it without one.

How do you figure that?

[EEVblog]

Baaad foundamental friday.
No, solution, no teory, no whiteboard.

and the world didn't end! 

[millerb]

And please do a mailbag, you've alredy 2 sagan of mail to open.

I second that, it's fun to see all the weird stuff people send Dave.

[kkp]

500mV noise: By trying to multiply in my head, and failing to do a reality check afterwards. Somehow I was thinking of the gain as 100k, with a resulting bandwidth 20Hz, and got it wrong by another factor of 10.

With 20kHz bandwidth, the 22nV/sqrtHz is dominating, so that should be sqrt(20kHz)*22nV/sqrtHz*gain=311uV RMS..
(checked. sounds about the right ballpark too.)

[JasperNL]

My guess would be the output load. I can remember a similar problem in my constant current load project. I also used auto-zero op-amp. Can't remember the actual part. The offset was little bit out of spec. Causes by a small capacitive load at the output. A decoupling resistor fixed the problem but decreased to bandwidth.
Can remember it pissed me off and ditched the part Used a non auto-zero part instead and accepted the higher offset.

[SeanB]

In the appnotes they show a 470p capacitor across the top gain setting resistor ( funny enough 100k and 1k) to give a low pass filter, so it is likely that what is being picked up is the output noise, it does have noise spurs at close to DC and at 15kHz. The load will not be a worry, it does already have a 100k load, the gain setting resistor. Adding a low pass RC filter to the output may help in any case.