Ultra low-noise, short-term stable references

[branadic]

Hello everyone,

at and after MM2024, there is/was quite some discussion going on due to the Z10 presented by chuckb. While we have no clue what chuck is using as his short-term reference, I thought it is worth having a dedicated thread concerning that topic and maybe some of you have some ideas how to put together something similar.
Having a low-noise reference combined with a noisy long-term stable reference is something we have seen in some multimeters before (1N829 + LM399 IIRC).

So here is my first take on that. I remembered MM2022 and the SPC1550 I got from Rainer:

https://www.eevblog.com/forum/metrology/volt-nut-meeting-2019-in-stuttgartgermany/msg4409104/#msg4409104

While noise already looked good, it was time to test stability of two SPC1550 connected in series and measured with a 3458A. I started a first measurement and I quickly noticed a huge sensitivity to ambient temperature changes (AC in the lab turning on and off more or less frequently), so the measurement was stopped and the SPC1550s put into a styrofoam box with TEC-controlled, temperature-stabilized conditions inside.

Attached are the data for that measurement as well as a zoomed image of the temperature stable area (roughly 4 hours) at the very end of the first diagram. While the short-term noise looks rather good and most likely limited by the 3458A, there is a very low frequency change visible.
So the question arrises, where is this coming from? To me it seems like a sensitivity to ambient pressure, but with a certain time constant. Could be the SPC1550s or partially the 3458A, which would mean the cells have to go into a pressure tight box to surpress the effect on them.

What is your take on that experiment?

-branadic-

[Kleinstein]

I would expect a super cap to behave a little similar to a battery. So there is a temperature dependence, depending on the chemistry and possibly charge history.

Another point is a possible effect of mechanical stress, e.g. from air pressure or via the leads and longer time humidity effecting plastic insulation.
AFAIR there are reports about the pressure effect of 10 V reference (more like LFLU based).

For the test one may have to measure the difference to a more stable 7 V reference, or use 2 such references and look at the difference.

[dietert1]

I'd guess it is about making a constant temperature, constant humidity and constant pressure oven. I think that was the claim in the Z10 info.
Making a constant humidity oven is quite difficult. Most ovens have some tiny leakage. So that results in a low pass filter behavior. For the IP66 enclosures with rubber seals i am using i have seen time constants up to 5 years. Of course there are much better and much more expensive hermetic enclosures and feedthroughs.
After reading the Z10 info i started to prepare a study including pressure measurement, using a Weidmüller K41 enclosure that we had. Results can be expected after a year or so. Probably there will be a low pass behavior also for pressure.
I know it sounds strange to write about years in this thread, but at "short" time scales low pass filters work very well. During my LTFLU studies i learned that for precision references what you see as noise at time scales of minutes to days are ambient variations attenuated by some low pass.

Regards, Dieter

[branadic]

The two SPC1550 I have today made their way into a sealed die cast aluminum case with sealed connector. Not fully pressure thight though, but the best I had at hand to see if this approach is worth to follow. The case went into the temperature stabilized styrofoam box. Tomorrow I have more data to show.

-branadic-

[chekhov]

IMHO, whatever makes Z10 reference wonderful is not the ultra low noise but the way how this low noise ref is adjusted or kept long term stable.
Let's imagine we have low noise source already, short time stable. And we have our array of LTZ/whatever long-term reference.
How to compare them, quickly or during sufficient time period, keeping in mind all 1/f noise produced by it, different fluctuations, popcorn, ...
How we adjust our low noise ref ... so that this does not affect the process of watching it as well

[iMo]

The Z10 box weights 50pounds (based on the preliminary DS), so there is a lot of metal inside - like a larger chamber or system of chambers, perhaps evacuated and ovenized (or cooled) of course..
The box could be 5kg, transformer and batteries 8kg, capacitors + electronics 5kg, so 5kg is the chamber(s)..
"14Ah 24V for 10hours of normal operation" may mean aprox 1.4Amp current, say 1.1A/24V (25W) for heating or cooling..

PS: my bet he is using large banks of low leakage capacitors (like 100 pcs. each) for signal filtering (or processing when using some servo loops), perhaps ovenized and shielded as well.

PPS: "4days transport power" - it could be he is switching off the ovenizing of the capacitor banks during the transport, leaving only the reference chamber ovenized (like 2-3W).

[branadic]

As promissed, here are results with the SPC1550s sitting inside a sealed Rolec case with sealed D-sub connector, while the Rolec case itself was sitting inside a TEC stabilized styrofoam box.
Obviously, things haven't really improved, but hey, that's why I made this experiement in the first place. 

-branadic-

[Kleinstein]

A low pass filter with some suitebale capacitors (e.g. supercapacitor) could indeed be a solution.  The capacitor seems to have quite some temperature effect though. To get an idea on how stable the temperature needs to be it may make sense to measure the sensitivity with some intentional temperature excursions. It can still be tricky with capacitor internal DA relatived effects that react to the rate of temperature change.

Another point that can make sense is a really sensitive temperature sensor (e.g. 10 K NTC) inside the stabilized box to see how good the stabilization works.

The way such a filter would likely work is to no expect DC precision from the fitler, but more use an extra comparison of the filter output (can include the buffer) and the long term stable references. The filter part is more like to slow down and filter the correction signal. The feedback part could in theory be digital, e.g. to incluce corrections (e.g. for long term drift or pressure).

[branadic]

Just for completeness, the TEC controller already uses a 10k thermistor to stabilize the temperature inside the styrofoam box to 23 °C.

-branadic-

[dietert1]

I think the logs in SPC1550-5.png exhibit a clear correlation of battery voltage with the sensor logs outside of the chamber and besides that very low noise. So the batteries with the 3458A are a good enough sensor to detect variations inside the chamber. In other words that chamber isn't good enough.
If the 3458A was running outside of the chamber, one could try and put it inside.

Regards, Dieter

[Kleinstein]

For really judgeing the temperature stabilty a separate sensor, not the one used for the regulation should be used. Ideally there would be not much temperature gradient, but one never knows before measuring it and the sense circuit to the NTC may add noise / drift that the regulator could cancel.

One way to get temperature fluctuations, especially with relatively large air volume is from pressure changes that convert to heat sources in air.

[branadic]

The styrofoam isn't large and the TEC controller with 60 W not strong enough to handle the heat produced by a 3458A.
Our large climate chamber is currently in use, so I have to either stop that pathway at the moment or test the reference at home, with way lower temperature excursions over a day, but larger overall temperatures at the moment.

-branadic-

[branadic]

I today hooked up the SPC1550s to my meter at home (higher temps in the room, but not corrupted by any AC). After an initial run down from equalizing to the new environment and some settling region I can now see the cells slowly discharge, but other than that it is low noise. 
Attached are the data (about 5 hours of data) in case you want to analyze them yourself.

-branadic-

[tszaboo]

Before spending a lot of time investigating and compensating all the physical effects, I would suggest to replace the supercap with one made by Tadiran.

[miro123]

  As already mentioned - used high quality supercaps.Select the one from low current range
  Why not place two caps good thermally couples and  with tick mechanical cushoning. Log temperature atmospherics pressure and vibration/accleration sensors./ Measure the voltage difference. Os such way you avoid many factors. Measure the temerature at least at to corners of capacitors. Temperature probing in non moving air is an issue. Alternatively you can use oil bath. I bought 1l. silicon oil for this purpose.
  1. The two capacitors has similar conditions they must drift on the same way and difference will less vary from outside conditions
  2. Difference should be in mV range. It will lessen the DMM/Nanovoltmeter or acq system requirements.
 

[dietert1]

branadics last log demonstrates a useful time constant of about an hour, e.g. between 19:30 and 20:30. Probably the diagram in the Z10 info isn't 80 minutes by accident.
The test also demonstrates that a battery can't be a reference but a filter for a reference.
With the usual 0.1 to 10 Hz spec of reference parts, we are looking at time constants of about 3 seconds. If a battery filter can extend that about 1000 times, then the filtered reference noise should be down about a factor 30. Let's say from 200 nV rms to 7 nV rms - approaching the Z10 1E-9 claim. I think branadic is on the right track.

Regards, Dieter

[branadic]

Rainer presented "supercapacitors as voltage source" at MM2022. In there he presented that EVE SPC1550 is similar to Tadiran HLC-1550A. So without further information of insights or experience a statement like "Before spending a lot of time investigating and compensating all the physical effects, I would suggest to replace the supercap with one made by Tadiran." is just like saying, a rocket to go to Mars has to be red. 

-branadic-

[Kleinstein]

The different bands likely use a very similar chemistry (if not even comming from the same source) and will thus have similar charactristics.
It can still make sense to use a major brand, as the chances may be a bit higher to sill get that type in a few years. So not so much because of the performance, but more a supply issue.

There is a chance that the properties depend on the actually used voltage and maybe the history. The 3.5 V performance may be different from 3.7 V. Also the history may make a different. I would be a bit careful with charging to a much higher voltage, especially close to the maximum, as this can alter the cell. There is a chance to get Li trapped deep inside the electrode and overcharging is suspected to cause cracks.

[iMo]

..
The test also demonstrates that a battery can't be a reference but a filter for a reference..

The power consumption of aprox 2-3W in the "transport mode" against aprox 25W in the "operational mode" supports that as well (see my post above). The reference itself is a small box of size of an larger OCXO perhaps (2-3W always ovenized).. The banks/chambers with the capacitors are much larger, ovenizing of which is switched off during the transport..

[miro123]

..
The test also demonstrates that a battery can't be a reference but a filter for a reference..

The power consumption of aprox 2-3W in the "transport mode" against aprox 25W in the "operational mode" supports that as well (see my post above). The reference itself is a small box of size of an larger OCXO perhaps (2-3W always ovenized).. The banks/chambers with the capacitors are much larger, ovenizing of which is switched off during the transport..

Small remarks. I see mixture of different thinks together. Battery Capacitor chargeble or not. Fanso Spc1550 is non-rechargeble LiSOCl₂ (Lithium Thionyl Chloride)  battery, There dozen of different super caps and chargeble anon chargeble batteries. All has different properties and application area. moer info over SPC 1550 form manufacture website - https://www.fansobattery.com/?list_46/516.html
@ branadic - Your setup shows limited capability. Electrochemical noise in batteries must be visible. It is not ultra-low noise either. I term of low electrochemical noise the Weston cell remains by far the best is this class.

[tszaboo]

Rainer presented "supercapacitors as voltage source" at MM2022. In there he presented that EVE SPC1550 is similar to Tadiran HLC-1550A. So without further information of insights or experience a statement like "Before spending a lot of time investigating and compensating all the physical effects, I would suggest to replace the supercap with one made by Tadiran." is just like saying, a rocket to go to Mars has to be red. 

-branadic-

Or maybe it's because Tadiran has decades of experience making batteries, and they don't cheap out on the assembly processes. They might look the same, the chemistry might be the same. And one uses laser welding to close, the other sticking the empty hole with some random resin, that changes every 6 months of production.
There are a lot tiny differences in batteries that you cannot tell from the outside.

[iMo]

..
The test also demonstrates that a battery can't be a reference but a filter for a reference..

The power consumption of aprox 2-3W in the "transport mode" against aprox 25W in the "operational mode" supports that as well (see my post above). The reference itself is a small box of size of an larger OCXO perhaps (2-3W always ovenized).. The banks/chambers with the capacitors are much larger, ovenizing of which is switched off during the transport..

Small remarks. I see mixture of different thinks together. Battery Capacitor chargeble or not. ..

Afaik all small size "super.." capacitors with XXX Farad "capacity" are built as batteries with chemicals inside. I bought one of the first Seiko Kinetics in '96 in Schiphol airport because it contained a supercapacitor thus no danger of leaking and no replacement of it needed. A big BS..   
I would never have considered to use those supercaps in a voltage reference which shall work fine for say 10 5 years..

[Kleinstein]

The limited lifetime of the chemical cell type capacitors could be an issue. If made right the cell can still work for quite some time. Using them at a stable and not so high temperature could prolong the life. So peltier cooling instead of just a simple oven could be a good idea.

[Doctorandus_P]

On the milli volt level, battery voltage is very sensitive to temperature changes. And the part you have chosen is not a capacitor, but a "Li-Ion battery with integrated capacitor". And I suspect that even standard electrolytic capacitors are temperature sensitive Materials expand when heated, and this changes distances, and thus capacitance. Same electric charge but with a different capacitance results in a different voltage.

Just like ESD. When you rub some materials together, some electrons jump to the other side, and when you increase the distance, the voltage goes up into the kV range.

Marco Reps has shown the temperature dependence of a regular 9V battery by just holding his hand near one. It's quite impressive, and enlightening too.

[tszaboo]

Marco Reps has shown the temperature dependence of a regular 9V battery by just holding his hand near one. It's quite impressive, and enlightening too.

If you have a 8.5 digit meter, you can change the reading by squeezing the battery, or holding it upside down.
I don't have that now, but a 34465A and a LiSOCL4 that I had handy, jumped 100uV just by me holding it in my hand for few seconds.