Ultra low-noise, short-term stable references

[iMo]

Those supercaps may work in RC filters for short-term measurements - when thermostatized and kept mechanically stable (like an at least partially evacuated chamber). With time their capacity (in Farad) will drop down, but that may not have large impact on the noise characteristic provided filter's sufficient low time constant will be planned ahead (to account for the expected capacity drop)..

[Kleinstein]

The temperature dependence is definitely a thing with battery cells. How much depends on the chemistry. With some  (e.g. non saturated mercury ref. cells) one can be quite good, but other may need a really good oven. Measuring the effect may be a first thing to see if there is a chance to reach a stable enough temperature.

Electrolytic capacitors also show a temperaure effect - touch a capacitor and the voltage tends to change. It can be more tricky than just constant charge and variable capacitance, as there is a high level of DA (5-10% range) and the history and not just the current voltage can have an effect. Similar for the battery cells it can make a difference whether it is fresh charges or coming from a high voltage before.

Because of the relatively large leakage one would likely not use a simple classic RC filter, but more like an active fitler of higher order. Here the higher order part could be implemented digital to a large part. This way a stable leakage or required charging current would not cause an offset. The question is only on where to put the cross over from the long term stable (e.g. LTZ1000) reference to the shorter time filter. This may have to be a rather long time constant, already in the region where the Allan variance curve of the final circuit will go up from slow drift parts. It would still need a rather good long term stable reference and thus very good design (e.g. stabile temperature also for the resistors at the LTZ ref.) there. With the filter I don't see an improvement much faster than ~  T maybe even only ~  sqrt(T) towards shorter times.

[moerm]

First: kudos to branadic, Kleinstein and some other gurus here (I mean that sincerely) for their work, knowledge and experience!

I'm a modest and pragmatic guy and am not even considering going beyond 1 ppm (with meters) and hence I'm not really interested in anything beyond honest and true 6.5 digits, which just so happen to be 1 ppm - but still I'm reading with attention and interest almost any and everything certain users (like the above mentioned) do and write about ... and learn from.

Here is my personal take on this topic:

I studied the ADR. Carefully. But I stick with the LTZ. Yes, in a few parameters the ADR is more attractive, noise being one of them. But for me, and presumably most hobbyists and people outside of cal labs, the "problems" the ADR solves are, excuse me, in between luxury and irrelevant. Why?
What I (I mean 'we' but I do not want to arrogate to speak for others) really need is long-term stability with acceptable to good precision and accuracy. Simple as that. And in that the LTZ still is unsurpassed. In addition it is well known and understood and "battle-proven" over decades.

Being in that department let me address another issue: I quite like the diverse long-term studies some of you made. But I don't care. At all. For two main reasons, a) Fluke, Datron, etc. make/made a living building the very best possible/feasible, and I guess trying to do better than them is quite futile (for mere mortals without a very serious budget), and b) how to measure (as opposed to guesstimate) and verify it? With an '8.5 digits' multimeter? Haha, good joke.
Btw, just look at that "new" (well, not really old) brit company "time" (or similar) who seems to be somewhat of a wannabe Datron "new edition" - and who seemed to fail and still be failing to achieve what Datron achieved decades ago.

Plus, again, IMO our problem isn't a few uV more or less noise. Our problem is 1 ppm per year stability.

Where I see the ADR is in places, for which the LTZ was overkill but LM399 isn't quite up to par that is, 7.5 and better 6.5 digits meters and not-quite-top reference boxes.

Again, thanks for your work. At the very least it helps us to better reflect those matters.

[Echo88]

Short term stability is irrelevant guys, lets end this thread, the 9.5 digit DMM thread and our hobby.
I wouldnt have guessed that youre modest, but after claiming it in the second sentence it must be true.

Anyway: Attached is a proposition for the supercap/ref/buffer.
The chamber parts are on the way.
At the moment i would stuff two ref/supercap/buffer circuits in the chamber and then intercompare the achieved noise via a 34420A with and without buffer.
The ADR1399 is just used as a cheaper alternative to the LTZ that would be used later, when this experiment actually works.
Any ideas what to improve/any found errors?
I assume the AZ-OP averaging will pose no problems with the supercaps, but i could be wrong here and i assume the AZ-OPs have a central clock thats used.
The hybrid supercaps were chosen for max capacity in the given chamber size at acceptable price point.
Mouser and Digikey stock no rechargeable Tadiran batteries, so they werent yet considered.
Since the chamber is very sturdy i didnt yet include a barometer, desiccant will keep the humidity constant and the chamber temp will be regulated by a peltier element, with the chamber in a dewar vessel.

[moerm]

Short term stability is irrelevant guys, lets end this thread, the 9.5 digit DMM thread and our hobby.

That's not what I said. Of course short-time stability, or more precisely, short-term accuracy and precision is important, too, but neither the ADR nor the LTZ nor the LMx99 mainly are designed for and serve for that. They clearly are meant for long-term stability and actually come with ridiculous initial accuracy.

As for ending "this thread, the 9.5 digit DMM thread and our hobby", how about the alternative of accepting views one doesn't agree with or like?

I wouldnt have guessed that youre modest, but after claiming it in the second sentence it must be true.

That statement was made in a context ("only" 1 ppm) and looking at discussions striving for single digit ppb desiring no more than 1 ppm indeed is modest.

Also, more of a side note, kindly note that neither did I attack or ridicule you ("ad hominem"), nor did I suggest that everyone should follow my view. In fact, I expressly thanked some users who seem to have a quite different point of view.

[dietert1]

A good bootstrap resistor for a ADR1399 from 10 V is 910R for about 3 mA current.
We have a set of 500F 2.7V supercaps mounted on a voltage distribution board. I measured the discharge current of that circuit to be about 1 uA at 27 °C ambient temperature, so that can't be used. I think between the voltage divider R2..R4 in your schematic and the capacitor nodes could be pairs of antiparallel diodes. Of course the divider should be fed from +10V_Ref (in front of R14).
Also the circuit needs some active control of total capacitor voltage to eliminate the voltage drop along R14 caused by capacitor leakage. I think there was a circuit proposal how to do it in this forum, but don't remember any details.

Regards, Dieter

[Echo88]

Indeed the modest comment wasnt seen in context, for that i apologize.
Lets agree to disagree, because our views differ here in many ways.

[Echo88]

Thanks for the suggestions.
R11/12/13 havent yet been fully determined; i still need to see if i have fitting resistors for the task in the drawer.
The antiparallel diodes between the cap voltage equalizing resistors and the caps is a nice idea.
Indeed the voltage drop over R14 could cause issues, so far i hope that the leakage remains stable, so as not to influence the voltage stability.

[moerm]

Indeed the modest comment wasnt seen in context, for that i apologize.
Lets agree to disagree, because our views differ here in many ways.

No problem, Echo88. That's what forum are for; to see, and potentially learn from, different views and angles.

That you apologize honours you but I didn't even expect it (but, of course accept it with a friendly smile).

[dietert1]

With 3x 220F the resulting capacitance is 73 F and for a time constant of one hour R14 isn't 10K but 50 Ohm. A different beast!

[iMo]

..and the 1399 wants the RC snubber..

[Andreas]

Hello,

the fuse F1 has a considerable resistance.
I would want to have it outside the filter cirquit. (at least not as additional ESR to the capacitor)

with best regards

Andreas

[iMo]

With 3x 220F the resulting capacitance is 73 F and for a time constant of one hour R14 isn't 10K but 50 Ohm. A different beast!

What should be the time constant actually?
ChuckB in his opamp thread indicated some opamps have good ADEV (below 1nV, ADA4528 and OPA189, the 4523 might be even better) till 1 day, afaik..
https://www.eevblog.com/forum/metrology/low-frequency-noise-of-zero-drift-amplifiers/msg2294556/#msg2294556

[Echo88]

Added the suggestions.
Thanks for the reminder of the OP-tests iMo, i forgot to look into it before going with the OPA4189.
Lucky that it turns out to be very good regarding stability and allows for 36V max supply, unlike the 5,5Vmax ADA4528.
Pdf and Kicad 8 Project are attached.

Maybe its beneficial to add an analog switch/OptoFET to change the R14 lowpass resistor inside the chamber, to check for different Allan Deviation? 

Edit:
Added layouts: main-pcb, stacked on supercap-pcb and thermal sensor alucore-pcb which is attached to the chamber.

[dietert1]

These Lithium ultracaps have roughly 100x lower or even better leakage than the standard 2.7 V ultracap. Looking at the AHC-S04R0S datasheet the leakage spec doesn't scale with capacitance but with square root of capacitance. If that's true one should use the 850 F instead of the 220 F part.
By the way i read some serious warnings about handling these capacitors as their fire/explosion risk is similar to Li-Ion batteries.
At 10 V the 3x 220 F cap takes 730 Cb of charge, that is 730 seconds of 1 A constant current charging. With a resistor it will take forever to approach 1 ppb.

Regards, Dieter

[Echo88]

Yeah, i was sceptical about lithium hybrid supercaps at first and had fears about degradation, the battery like chemistry and the limited discharge voltage (opposed to no minimum dicharge voltage will nonhybrid supercaps).
But the about 10-fold capacity increase and the very low leakage currents are great, so i will try them.
The chosen 220F caps are used because of chamber size limits, otherwise i would have gone for the available 850F versions.
I also have a bit of fear about accidentally shorting those babies, thats why i tried to limit/fuse all outgoing cap connections, while allowing for lowest noise measurements.
Youre completely right about the charging taking forever even with the low resistance fast charging path. Would spoil the fun when it takes some years to stabilize.
Added the new schematic: current limited source instead of simple resistor and protection measures against the charged supercaps when no +18V supply is applied.
Dielectric absorption is another worry: https://www.doeeet.com/content/wp-content/uploads/2022/12/Table-4.-dielectric-absorption-of-most-common-dielectrics.png
Am i correct in assuming that R14  / Q3 Beta = Effective R14 for the lowpass consideration?
-> 1k / ~100 = 10R low pass resistor equivalent?

[Kleinstein]

For the output driver there is more than R14 / beta. One thing is that R5 adds to the resistance. The other part is the effective emitter resistance for the transistor. Here 26 mV divided by the current for Q3 is the usual approximation.  This could become an issue, as the current finally will go very low and thus high effective resistance.

The chain for balancing the capacitors would act as a rather missive leakage path. From the logic R2,R3,R4 should be more connected to the output buffer. However this would cause problems with the off case and would need extra care (e.g. extra switches or similar).

For a proof of concept there is no need to use 4 OP-amps in parallel for the output buffer - the extra current and power consumption there can also cause problems.

For U3 one would more look at an OP07 or OPA189 instead of the OPA140.

The ADR1399 inside the same thermal chamber would be quite some heat source. Normally one would want to avoid heat around the capacitors, as leakage goes up and it would take a rather long (e.g. maybe a day) time for stabilization.

[Echo88]

I hope i understood your statements about emitter resistance and the balancing resistors R2/3/4 correctly.
The emitter output resistance can be kept low by allowing a constant 1mA (roughly 26mV/1mA = 26R?) to flow through R2/3/4 which i decreased in value and changed their groundpath, to omit this gnd-current from the supercap star gnd.
Is that okay or does this 1mA need to flow into a separate negative supply instead of gnd, like in this example, see figure 22?: https://www.analog.com/en/resources/technical-articles/simulation-shows-how-real-op-amps-can-drive-capacitive-loads.html
Seems to be lowest part count solution to this problem.

Another solution might be a different limited current source or the source (R14,Q2,Q3,R5) could be bypassed by a resistor via switch after the supercaps are sufficiently charged.
I changed the OPA4189 to a OPA189.
U3 OPA140 being a SOT23-5 footprint can be directly swapped with a OPA189 or OPA207 for example.
I plan to have little thermal baffles around the ADR1399 and since the whole chamber is thermally stabilized im not really worried about temp fluctuations inside the chamber once the supercaps are charged.
The changed schematic is attached.

Thanks for your contributions 

[Kleinstein]

Driving the super cap with a low impedance is a strange way, but some 50 ohm resistance is needed to get a time constant of around 1 hour.
Still the 10 K resistance from the balancing part looks odd and give quite some "leakage" compared to the supercap.

With the capacitive drive circuit the capacitor C7 may also need to get quite large. So C7 gets nearly as critical as the main capacitor, with problems from DA and so on.

[Kleinstein]

A possible solution could be replacing the driving OP-amp and filter part (U5, C7,R17) with a digital solution. So amplifier, ADC, µC, DAC and likely a current source. The digital solution can be really good in integrating and handling the slow part. For a digital filter the rather long time constant is not a problem: The demand on the ADC and DAC should not be that large, as this is mainly for a tiny difference (possibly even just some 10 µV) and thus not much resolution needed to reach the limit of the reference noise. One could possibly even get away with a µC internal ADC, though likely want a better DAC.
Chances are one may want some range switching for coarse and fine approach.

[iMo]

The R14 (50ohm) looks to me too small. When you short the output of the Star+ to ground the output of the opamp goes full high and the opamp's output current (XXXmA) will flow through the 50ohm, and BE of the output transistor (or CE of the current limiter) and/or R5 to ground. I would put there 1k..