EEVblog #1057 - µCurrent Murphy

[EEVblog]

Dave investigates a problem with a recent batch of uCurrents, something that WASN'T supposed to happen!
A rare look at a real world combination of design, manufacturing, production testing, and procurement problems.

[NiHaoMike]

With the problematic opamps, what happens if you connect a 100n capacitor from battery negative to output ground? I think the lack of bypass from supply rails to output ground might have been a latent design flaw.

[IanB]

I notice they show out of stock in the shop. Did you withdraw them from sale while you investigate? What will happen to the ones assembled with the bad chip?

[2N3055]

I had the same problem...
I also didn't need high bandwidth, so I added some capacitors.

Basically rail splitter was oscillating, but it depended on power source. If battery was new or if I powered it with PSU it worked well..
It depended much on internal impedance of the source..

[EEVblog]

and

[EEVblog]

I notice they show out of stock in the shop. Did you withdraw them from sale while you investigate? What will happen to the ones assembled with the bad chip?

No, they just happened to be out of stock.

[TMM]

Connecting bypass caps to both rails is ill-advised as it creates a PSRR issue. Just one cap to either rail should be fine because you already have another bypass cap between the rails 

[blackdog]

Hi Dave,


Dit you read my remark in the other topic that the feedback capacitor of the last opamp is on the wrong position.
It should be direct on the opamp output and not on the output of the device.
You will create another pole if you load the output with "to much" capacitens and if the phase margin and/off te power supply rejection of the problem opamp is less,
you wil get a nice generator 

Kind regarts

[tszaboo]

I had a similar issue with simple jelly bean TVS diodes. SMAxxCA. The originaly used, ST part had 1/10th of the leakage current at room temperature, than the replacement part from Taiwan semi. A few hundred board had to be reworked.

[eV1Te]

Hi Dave,


Dit you read my remark in the other topic that the feedback capacitor of the last opamp is on the wrong position.
It should be direct on the opamp output and not on the output of the device.
You will create another pole if you load the output with "to much" capacitens and if the phase margin and/off te power supply rejection of the problem opamp is less,
you wil get a nice generator 

Kind regarts

This was also my first thought when I saw the schematics, the feedback capacitor for the last OpAmp goes directly to the output where you have an unknown amount of capacitance. This might not be the actual issue in this case but I always try and add a final resistor for some isolation since you never know what it will be connected to.

For example, what if the DMM/circuit you are connecting to actually outputs some voltage/noise, will that be coupled back into the feedback circuit?

In this case I would make a quick test and add a 1k ohm resistor between the final output of the microcurrent and the DMM you are connecting to and see if that helps. Or move the feedback network to the OpAmps output directly, but this will require soldering. 
The increased output impedance should not influence frequency response unless the capacitive loading is getting close to 1 nF.

[MiDi]

Would suggest it is mainly a CMRR/PSRR problem of the first amplifier stage (second only distributes 1/10).
A look into the Datasheet of MAX4239 for this turns out that at 1Mhz the CMRR or PSRR are in the order of 0dB rejection (at higher frequencies even worse? Can it go below 0dB rejection?).
A CM or PS-ripple voltage at this frequency will be magnified by 110 by the amplifiers, this is more than input signal amplification!
1MHz 100µVpp on the supplies or the VGND turns into ~11mVpp at the output.
A filter for CM/PS for high frequencies on first amp stage should improve the circuit.

A measurement of the CM and PS-ripple on first stage would be interesting.

[ogden]

A measurement of the CM and PS-ripple on first stage would be interesting.

Right. I vote for thorough stability measurements and analysis of whole circuit. Last stage is especially interesting from stability point of view. Could be so that problem is complex - last stage instability together with inferior CMRR/PSRR of rail splitter, both creating some feedback circuit. "Opamp stability" and compensation could be good topic for video.

http://www.ti.com/lit/an/sloa020a/sloa020a.pdf

[jimmc]

I think that the situation is not helped by the placement of the feedback capacitor C4, it would be better to connect C4 to the Op-Amp side of R8.
At present R8 and the load capacity  introduce additional phase shift in the AC feedback loop via C4 reducing stability margins.

http://www.analog.com/en/analog-dialogue/articles/techniques-to-avoid-instability-capacitive-loading.html


I had a similar problem with a LEM PR200 current probe where moving the feedback capacitors cured the problem.

Jim

Apologies to Blackdog, who already mentioned this in post #7.
I'm not having a good day  .

[coppice]

I've had bad experiences swapping between makes of the LMV321, LMV322 and LMV324. They vary in a number of ways. Like most op amps described as rail to rail, they don't actually work close to the rails. However, different makes vary considerably in how close to the rails they work. Stability into capacitive loads varies considerably, as Dave has found. They are certainly a go to part for low performance jelly bean op-amps, but you need to try your design with various makes if you intend to swap suppliers during production.

[analogo]

Which capacitance decade box is Dave using? I see it is marked with the open source logo, but I could not find it in the usual places.

[Neilm]

I have had similar problems with certain sensitive circuits, especially with transistors. The assembly notes for the board now read that only a specific manufacture and part can be used. It looked as if some parameters (not on the datasheet) were the cause.

[Tsippaduida]

I haven't run into this issue with my uCurrent, but I checked it and it has the "wrong" chip also. My serial is ~428x, don't remember it now. I mailed t he information to Dave in the hopes it will help in narrowing down the affected units.

I might order replacement chip when I need to get more parts next time.

[bitwelder]

Searching on Youtube for uCurrent, I found this old video by electronupdate, where he puts the LM321 die under the microscope... and finds a Fairchild labeled FAN4114A. Interesting.

[BU508A]

Which capacitance decade box is Dave using? I see it is marked with the open source logo, but I could not find it in the usual places.

I think it is this one:

https://www.ebay.com/itm/182642398119

[ckambiselis]

Maybe I haven't seen enough chips in my life yet, but every time I see a chip/fet/... with that glossy straight cut edges plastic package it's a fake Chinesium chip I have bought from eBay for some quick testing, otherwise all I get from Farnell, Distrelec, TME etc. have the matte finish rounded corners plastic package.

Waiting for some FETs by ON Semi to come from Farnell, lets see what the package looks on them, maybe it's how ON Semi makes the chip packaging.

Oscar

[Dave]

Here, this should fix your design flaws.
There are two caps filtering the rail splitter reference, so the voltage jumps directly to the midpoint at startup instead of slowly ramping there.

[TMM]

Stability into capacitive loads varies considerably, as Dave has found.

I don't think that was the issue here. There was a feedback path via the power supply and the phase shift of everything in that path made it positive feedback. Significant phase shift occurs in an opamp above it's operating bandwidth. Makes sense as the opamps used are ~1MHz GBW and the oscillation was at 2.7MHz. Some variance in the GBW between opamp versions would mean that under certain circumstances some opamps (higher GBW than spec) are stable and others (GBW near spec) oscillate. 

Here, this should fix your design flaws.
There are two caps filtering the rail splitter reference, so the voltage jumps directly to the midpoint at startup instead of slowly ramping there.

You don't want to do that because the two caps work as a capacitive divider and any common-mode noise on the rails couples through. Using just one cap gets you an RC filter. There is already a decoupling cap from +rail to -rail so by adding just one cap from VGND/R7 to either rail there will be a high frequency path from the decoupled node to either supply rail. The decoupling caps probably only need to be small to solve the oscillation issue (a few hundred pF will probably do) so there will be no noticeable turn on delay. Even 1nF gives a turn on settling time of a few milliseconds.

[fusionmkx]

Could you share where to get that the variable capacitor and variable precision resistor devices that you've been using??

[BU508A]

Could you share where to get that the variable capacitor and variable precision resistor devices that you've been using??

variable capacitor decade box: see five postings above. Just scroll up.

[bitwelder]

Could you share where to get that the variable capacitor and variable precision resistor devices that you've been using??

Search for Atlanta Robotics decade capacitance box.
http://www.inmojo.com/store/atlanta-robotics/item/decade-capacitance-box/
(the page includes the design in Eagle if you want to build it - it's OSHW - and a link to Dave's review a long time ago)