Ultra Precision Reference LTZ1000

[Noopy]

I expect that AD generally treats any changes on any component following the Automotive rules, as their QA processes will not distinguish between the commercial and automotive business.

The companies have to invest some money and manpower to fulfill the automotive requirements. I don´t think they do it for all their parts.
They won´t have totally different QA processes but if you can cut corners without getting problems...
(In fact I assume it is the other way round: The consumer QA process is the baseline and then you put additional requirements on top of it if you have an automotive or military part.)

Are you also working @ Automotive?

Yes, for 15 years now. Are you in this area too?

[Dr. Frank]

I expect that AD generally treats any changes on any component following the Automotive rules, as their QA processes will not distinguish between the commercial and automotive business.

The companies have to invest some money and manpower to fulfill the automotive requirements. I don´t think they do it for all their parts.
They won´t have totally different QA processes but if you can cut corners without getting problems...
(In fact I assume it is the other way round: The consumer QA process is the baseline and then you put additional requirements on top of it if you have an automotive or military part.)

Are you also working @ Automotive?

Yes, for 15 years now. Are you in this area too?

Yes, since 29 years: First and shortly @ SQM for electronic parts in manufacturing, then in R&D, Component - Management, - Technologies, PCN, PTN, R&D processes, PLM, Maturity processes for components and products, Robust Design, and so forth.
Actually, the R&D assessment part of the ZVEI matrices were developed in my team, 20 years ago, and I'm still working on that PCN/PTN stuff.

Suppliers (Tier2's) usually stick to Automotive rules for all of their components, as soon as they deliver only one device to us.

Frank 

[Noopy]

Oh, cool!
I did a lot of R&D stuff but always hardware related, trying to avoid software wherever it was possible. 
Since 2,5 years I'm now in our Robust Design department: consultant for R&D, (schematic and layout) reviews, hardware change management, validation, failure analysis, obsolescence management, supply security (the technical side),...
Who knows perhaps we will accidentally meet. 

As you know at work we never ever want to see non-automotive parts. I was told you can never be sure what you get if you choose non-automotive. For sure it's not just black and white.
...I will ask the next application engineer I meet. 

Richard

[Dr. Frank]

Hello,
I want to finalize the characterisation of my "travelling" standard, the LTZ #3.

It consists of an LTZ1000 reference @ 50°C oven, a 10V amplifier, and all resistors are BMF types from AE = Alpha Electronics.

I already measured the T.C. of the 7,131V directly from the LTZ chip, which is about -0.025ppm/°C (box method)




The T.C. of the 10V output, mainly determined by the A.E. amplifier resistors, is about +0.06ppm/°C, which is a great match, as this type FLCY has a nominal T.C. of about -1ppm/°C.




Then I monitored the timely drift of this ratio, by always measuring both the 10V and the 7,131V outputs during my transfer measurements.
I have no clue, why the curve is so bumpy.
Anyhow, if you have a close look on the 6 months from 1.1.24 - 1.7.24, the drift is about -0.5ppm, the same observation is valid for 1.1.25 - 1.7.25.
Therefore, I deduce that the annual drift is about -0.25ppm/y., which is also very good, because the resistors are "only" conformally coated.





Frank   

[Dr. Frank]

I checked again the ratio drift history.

The dip in Nov. 23 was due to the re-assembly of the LTZ#3.
I added an outer case and made some other changes on the assembly and circuit, but measured the outputs directly after.
The LTZ #3 direct 7.13V output showed no change at all, but for the 10V output, the handling obviously influenced the trim pot.

In August 2024, I took the LTZ #3 on a trip to the Metrology Meeting 2024 in Stuttgart by a 2h car ride.
The vibrations obviously as well affected the trim pot and changed the output of the 10V again, as the 7.13V output was not affected by the ride at all.

For these 2years, I finally add the absolute drifts of the 7.13V and 10V outputs.
The 7.13V drifted by -0.5ppm/y, and the 10V by about -0.7ppm/y, which confirms the -0.25ppm/y drift of the ratio resistors.

In summary, I could well rely only on the 10V output, providing the reference would not be moved or shaken.

For exchange of the Volt with other volt-nuts, this 10V reference would be stable enough to make a transfer on a <1ppm uncertainty level, or within 0.2ppm, if a ratio transfer standard (like the 720A) is available on the receiving side.

Frank

[Dr. Frank]

Here's my schematic for the 7V => 10V amplifier, the initial design goal was to find an easy way to use standard values for the coarse and fine trim resistors, and to avoid custom made Vishay precision resistors.
I also expected a big annual drift of R12 and R13 on the order of 20ppm/y each.

These R12=10k and R13=4k FLCY resistors were offered as standard values, and the fine trim resistor R14, a thin film type, could always be chosen from a standard E96 range.
That left 480ppm trim range for the trim combination R15 || R18, in a way calculated so that the trim pot could just be adjusted to 0.5ppm at its resolution limit.

I have to admit that this is really a violation of Robust Design in favor of easy component selection.
Of course I was aware that in all old FLUKE gear, such trim pots would always have been selected for a trim range to cover only the lifetime drift of the reference, i.e. about 20ppm trimming range.   
Really careless thinking and a real B.S. design decision.. but at that time, this aspect was really not important to me, as I was completely focused on the expected stability of the LTZ1000.
For maintaining the Volt,  the 10V amplification was just an unnecessary add-on.

Therefore, I will now change R18 to 1 Ohm to reduce the trim range to 25ppm, and increase R14 by the difference of the trim combination.

That will give an absolutely rugged and stable 10V reference.
If I would equip all my 7 other 7,xxV references with these proven timely stable resistors, I could sell my HP3458A  and do the regular comparisons of my 12 references with a simple Null voltmeter only.

Frank

[Andreas]

Therefore, I will now change R18 to 1 Ohm to reduce the trim range to 25ppm, and increase R14 by the difference of the trim combination.

Hello,

What about the wiper resistance?

I would rather use the "precision trim" from the LM399 data sheet.
Although I have not the best experiences with trim pots at all (changes after temperature sweeps).

https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg5683669/#msg5683669

with best regards

Andreas

[iMo]

I would use wire wound trimmers (even they are bulkier) as they have got "fine detents" actually. In Frank's case those 20ohm. You may get a plastic retention friction fit collars for them such the shaft does not move.
The plastic 10T trimmers have got plastic gear with some dead zone, my experience is when you adjust it in the middle of the dead zone the stability seems be better. Their resistive track length to wiper contact area is smaller here, so the wiper's resistivity fluctuation contribution will be higher.
AI tool says:

[Dr. Frank]

Hello,
 I woke up from my metrological Dornröschen sleep, and I have created a long, long bucket list of postponed projects, i.e. to redesign, improve, make or buy metrology instruments.
One of these is the improvement of the LTZ1000 reference board inside my hp3458A.
I already have set the oven temperature down to 65°C, so it showed a very small timely drift over 15 years of <1ppm.

I now added Andreas noise improvements by 5 capacitors and one resistor, see schematic.
Btw.:  I ruggedized my LTZ#3 with the original FLUKE trimming method, works great and gives 0.1ppm resolution with standard parts.

Then I doubled the zener current by changing to a 50 Ohm resistor.

Both measures decreased the StD and therefore, the uncertainty of DCV measurements down by a factor of 1.5 .. 2.
Before, I always measured my ADR1000 reference at typically 150nV StD, but rarely below 100nV, but now, ALL consecutive measurements are well below 100nV, but rarely @ 150nV, at most.

Last, I removed the hp plastic cap and put another cap all over the reference directly on the PCB. This already improved the StD of my 34465A by a factor of 1.5, as the KS engineers made this design flaw to let the fan blow directly over the reference area.
I doubt that it'll help here, as the LTZ does not sit directly in the air flow.

On next occasion, the reference will be replaced by an ADR, and the +/- 12V reference OpAmp by less noisy types.
Unfortunately, I did not get any feedback from TiN or MiDi, if this really would help.

Stay tuned, I really have a lot of intersting stuff in the pipeline, next thing will be chasing the very noisy reference return path of my DATRON 1271, which leads to very bad 500nV StD in comparison. The 8508A might have this design flaw as well, but unfortunately, I did not find neither DMMs on TiNs DMM noise project.

Frank

[iMo]

I would clean up the flux residuals on that ref pcb with IPA carefully..
I would do it before soldering the Wima caps in there, and after soldering in the Wima caps I would carefully clean up the the flux only at the caps soldering points such I would not pour the IPA over that Wima capacitors.

[Andreas]

Hello Frank,

I would put a 20-50 R isolation resistor (against capacitive loads) between IC3,6 + C10 and 10V output JP4 + R18 to prevent oscillations of the 10 V output with capacitive loads >1nF.

With best regards

Andreas

[MiDi]

On next occasion, the reference will be replaced by an ADR, and the +/- 12V reference OpAmp by less noisy types.
Unfortunately, I did not get any feedback from TiN or MiDi, if this really would help.

Nice to have you on the journey, especially as your unit has already very low noise w/o mods.
You can find the comparisons of each mod in my unit in the 3458A Worklog.
I had the chance to compare to another A3 and it revealed that the original A3 (U180) has quite high noise, so right now it looks U180 is limiting noise improvements - details follow.
If you need more details let me know.

[Dr. Frank]

Thanks both for your advices, but it's getting even better.

I guess I made an essential, ground breaking discovery and improvement on probably all existing long scale DMMs.
These all have an application flaw, but nobody before has noticed that, neither LT, nor any DMM design engineer.
This effect might only affect the LTZ based DMMs, as this following root cause might destabilize the oven control circuit only.

When making direct comparisons with this absolute method by means of an DMM, I always wondered, why I never could achieve the very same noise figures as the differential method, although both methods are essentially identical: Any DMM directly measures the difference between two (LTZ1000/LTFLU/LM399) references, and the A/D only serves as the differential instrument, contributing no additional noise.



I always observed 160nV @ 10min and 200nV @ 1h time constant.

Please observe these strange humps, very prominent after 33600sec, and being always present, unavoidable.
Please check this on your own stability measurements with your specific 8.5 digit DMMs.



I always assigned these to my specific LTZ1000A on A9, assuming that it suffers from sort of 'random walk desease', maybe being a different effect apart from the usual 'popcorn noise' of the LM399.

After Andreas ingenious invention, I could demonstrate, that his circuit has the effect of effectively suppressing any external ever so slight e.m.c. or similar disturbances.
These reference amplifiers are extremely sensitive, as it's commonly known.

The comparison with the improved A9 board now shows exactly the same noise behavior as the differential method, with the exact same noise levels for 10 min, which is a -33% noise improvement.
The low 1min noise can be assigned to the doubling of the zener current, giving another -33% improvement.
I estimated 50% in total, before.



This reduces my Type B measurement uncertainty greatly, making long term monitoring stable to the physical limit.
I as well always wondered, why all my transfer measurements of 15 references showed 0.2ppm wide fluctuations from month to month, although each sample can be made with 0.02ppm StD, and all references are as well stable to this noise level.



I bet, that this effect as well causes the instability in DATRONs 1271 reference circuitry. This part of the DATRON design is extremely complicated and over-engineered. John R. Pickering generally comfirmed to me over-engineering of his former DATRON engineers,which was the decisive competition disadvantage over the HP3458A development and although the 1271 is really much better designed.
Their reference sensing method, which is basically a very clever idea, creates an instability source with small oscillations directly impacting the LTZ1000, which would normally not create voltage reference instabilities. Btw, DATRON already used Andreas' low pass feedback 10k/22n on the amplifier.

As this sensitivity characteristics is inherent to the 'Reference Amplifier', probably all DMMs have inherited this problem, but  this can easily be fixed.

Frank

Andreas, you Wizard of LTZ, now I'm really happy to meet you for our small MM on Saturday!     

PS: for obvious reasons I'm just calibrating my HP3458A.
It came to my mind, that this is the first time ever, that it's now possible to make a proper calibration, w/o an error directly from the factory, however small this might has been 

[Kleinstein]

I kown it gets a bit off topic here. From my analysis there is not much extra noise from U180. The noise of the low noise units is about what one can expect from the architecture and essentially not extra noise not acounted for or for jitter of the U180 internal switches. So it's a limit of the ADC design, not U180. It is possible that some of the U180 are more noisy (higher jitter) than others. After all the reported noise does scatter quite a bit (e.g. 500 - 700 nV RMS for 1 PLC AZ shorted). The difference is quite large for just different levels of 1/f noise in the LT1001 (espcially the one at the integrator).

The amplifiers for the 7 V to +-12 V step add a little to the noise, but not really much. I don't think it would be worth messing with these.
The low hanging fruit may be the 100 kHz range noise of the reference. Depending on the ref. noise this may be well more than the LT1001 noise, but still not a huge improvement. The ADC should be somewhat sensitive to this and mix it down to near DC.
Filtering for the higher frequency noise should be relatively simple. With my ADC I have seen quite some effect from filtering the ref. noise - though starting with a much more noisy LM399.

The reference noise in the 100 kHz range can depend on the ref. circuit details (e.g. the capacitors). The specs I have seen also don't go that high in frequency. In the simulation some of the circuits show even noise peaking in this range. The mod's by Andreas may already help a bit with the higher frequency noise, but there is still the LT1013 noise.

Most DMMs in AZ mode only sample the input half the time and effectively also use there internal reference only half the time. This can effectively add to the noise BW for the internal reference and also the DUT.

[iMo]

I wonder how the 300 Hz chopping of the LT1052 on Frank's board might affect such delicate measurements. The blocking with those WIMA capacitors may have helped suppress its possible disturbances.

[Dr. Frank]

TiN, MiDi,
here's another final consequence out of these findings:
The 3458As A/D circuit, especially the two reference OpAmps was thought to add noise to the readings, this turned out NOT to be true, as the 3458A now responds exactly like the differential circuit, leaving no space for any additional noise sources.
This means, you only get the noise improvement by adding the ADR1000, i.e. another 50nV.

That's it with your investigation on this subject.

This means, these 3 measures will improve the noise figure by a factor of 2, which is really great.
Another conclusion: As the 3458A was always on par with all other LTFLU based instruments, this reference as well is affected by this irradiation effect.

I'm really feeling great having the only and best 8.5 DMM in the world with the lowest noise performance.


Frank
 

PS, here comes my very last conclusion on this topic.
This is a minor effect, i.e. only 0.01ppm. That's not metrologically spectacular at all, and it does of course not yet explain my 0.2 jitter in the monitoring figures. For that, I have found a better Transfer workflow with <<0.1ppm uncertainty for 15 references, currently done within 45min. I will report on that later.

But this is a small metrological instrument sensation, as now the PTB or the NIST can easily and comfortly use the 3458A as a transfer instrument down  in the tens of nV uncertainty range. One has only to assemble enough high current ADRs in parallel, what TiN already has tried out.

[Dr. Frank]

I wonder how the 300 Hz chopping of the LT1052 on Frank's board might affect such delicate measurements. The blocking with those WIMA capacitors may have helped suppress its possible disturbances.

The 1052 were used, because these are the only ChopAmps left with low bias on the order of 20pA.
I have meanwhile of course assembled ADA4522s everywhere, as the divider resistors were chosen low ohmic.
And no, the 1052, despite its noise, made no remarkable influence on the LTZ1000 at all.

Frank

[Andreas]

DATRON already used Andreas' low pass feedback 10k/22n on the amplifier.

Its the other way round: I copied the Datron reference design at this point because from previous experiences with EMI it was the only logical solution for me.
(Tested this also with a steel needle held in my hand contacting every LTZ1000 pin to find the most sensitive ones).

with best regards

Andreas

[Dr. Frank]

Ok, my final thoughts:

At first, this discovery makes all differentiations between 5, 6, 7, 8 digits DMMs completely obsolete or irrelevant, as all DMMs perform identical concerning useful resolution. The differentiation only consists of the INL, the stability of the components (gain), features and functions. Resolution is in most cases only a software question.

All the Keysight DMM 401, 411, 465, etc. perform exactly the same as the 3458A, apart from INL.

The 3458A resolves 9 digits.

Its non-linearity is 8 digits only, but non-overlapping, which can be linearized.

As this mysterious noise barrier now has vanished, the 3458A is in fact a 9 digit DMM, always been.

Frank 

[aronake]

Btw.:  I ruggedized my LTZ#3 with the original FLUKE trimming method, works great and gives 0.1ppm resolution with standard parts.

Guten Morgen!

Some great and not too difficult to implement tweaks!

What trimming method is this referring to? Resistor in parallelly with a trimmer to make trimmers impact less?

Then I doubled the zener current by changing to a 50 Ohm resistor.

Any idea how this would impact TC of the reference? What disadvantages are there with higher current?

I always measured my ADR1000 reference at typically 150nV StD, but rarely below 100nV, but now, ALL consecutive measurements are well below 100nV, but rarely @ 150nV, at most.

With 100 nplc or?

Last, I removed the hp plastic cap and put another cap all over the reference directly on the PCB. This already improved the StD of my 34465A by a factor of 1.5, as the KS engineers made this design flaw to let the fan blow directly over the reference area.
I doubt that it'll help here, as the LTZ does not sit directly in the air flow.

Why not keep the original cap also for some "double shielding"? Or the heater in the LTZ might help to stabilize temperature for the whole board?

The mentioned resistor to add for lower noise, which one is that? I only see the 100K added for lower heater temp, but did not look too carefully.

[Dr. Frank]

Btw.:  I ruggedized my LTZ#3 with the original FLUKE trimming method, works great and gives 0.1ppm resolution with standard parts.

Guten Morgen!

Some great and not too difficult to implement tweaks!

What trimming method is this referring to? Resistor in parallelly with a trimmer to make trimmers impact less?

Then I doubled the zener current by changing to a 50 Ohm resistor.

Any idea how this would impact TC of the reference? What disadvantages are there with higher current?

I always measured my ADR1000 reference at typically 150nV StD, but rarely below 100nV, but now, ALL consecutive measurements are well below 100nV, but rarely @ 150nV, at most.

With 100 nplc or?

Last, I removed the hp plastic cap and put another cap all over the reference directly on the PCB. This already improved the StD of my 34465A by a factor of 1.5, as the KS engineers made this design flaw to let the fan blow directly over the reference area.
I doubt that it'll help here, as the LTZ does not sit directly in the air flow.

Why not keep the original cap also for some "double shielding"? Or the heater in the LTZ might help to stabilize temperature for the whole board?

The mentioned resistor to add for lower noise, which one is that? I only see the 100K added for lower heater temp, but did not look too carefully.

Hello,
this trimming method overcomes the problem of having to trim very small values in the 1 Ohm range with much higher,available trimmer values like 100 Ohm. Often used by Fluke, in various configurations.

I did not test yet the impact on the TC, seems to be altered greatly. We'll see how to compensate this. Will be done with the ADR1000 anyhow, so this is completely new game. No disadvantges existt at all. See specification as well.

Yes, always NPLC100 to have a standard comparison.

I always found this cap useless, thermally and electrically, and all my references do not have a cap for the LTZ.
This thing does not make any difference in perfomance, but will influence / shift the overall TC. So what? No double shielding required. It made no sense to further dig deeper in this matter, as this more simple, non over-engineered thermal design fulfills the requirements of TC, drift as good as the cap version. These would be non or hardly detectable effects, orders of magnitude below the key drift characteristics of such a references. This is completely useless and esoteric stuff, like all these slots, rings and alike.   
What you cannot measure does not exist, and no further second to think to not waste any effort or money on..
First is one Physics rule, the 2nd is an Automotive Engineering rule.

Frank

[Dr. Frank]

As branadic asked, the additional capacitors by Andreas are: C4, C5, C6, C9, 4x 100nF, plus R9, 10k plus C3, 22nF (instead of 2nF only)


branadic also wondered, whether his PREMA 6048 would benefit from upgrading its LTZ reference.
Definitely, as it evidently as well suffers from a high noise level, i.e. they use these ultra long integration times only to get stable readings.
After adding these capacitors, I bet that he can use 2sec integration time for 8 digits.
Unfortunately, 20sec is the fastest setting, maybe PREMA can provide the Firmware to enable 8 digits for faster NPLCs.
Anyhow, this improvement should be well visible on the 7 digit integration times.

I'm not aware, which other 8 digit DMMs exist, with an LTZ reference and long integration times.

I will report back, when I finally upgraded my 1271.

Frank

[Kleinstein]

With DMMs there is more than just reference noise. From the little we know about the prema ADC the long integration time may be just needed to get enough resolution, e.g. reduce the quantization noise. So I would not expect much improvement in the ADC noise or speed from a better reference.

Making the reference more resiliant to EMI can still be a good thing. This is not only interference from extern, but also meter internal EMI. This can look like noise with a crystal clock for the computer part and a mains locked PLL for the ADC. RF interference between circuit parts can cause nasty surprizes (not just noise, but also linearity problems) in precision circuits.

[Nixfried]

I'm not aware, which other 8 digit DMMs exist, with an LTZ reference and long integration times.

Hi Frank,

hope you are doing good!
What about the bigger brother of your newly acquired 1271, the 1281?
It needs even more time, 25 seconds if i remember correctly. Also the Fluke 8508A comes to mind.
Yes is know, later on they changed to an LTFLU reference, but still...

Would it be okay to call you again? (If you don't remember me, I'm the PREMA and Valhalla guy from like almost 5 years ago!)


Tim

[Dr. Frank]

I'm not aware, which other 8 digit DMMs exist, with an LTZ reference and long integration times.

Hi Frank,

hope you are doing good!
What about the bigger brother of your newly acquired 1271, the 1281?
It needs even more time, 25 seconds if i remember correctly. Also the Fluke 8508A comes to mind.
Yes is know, later on they changed to an LTFLU reference, but still...

Would it be okay to call you again? (If you don't remember me, I'm the PREMA and Valhalla guy from like almost 5 years ago!)


Tim

Hello Tim,
of course I remember you, nice to read you again!

Due to the mediocre feedback here, I have to admit that I was way too euphoristic concerning this disciovery.
This is relevant or interesting for TiN, MiDi and PTB/NIST at most, as it's an 0.01ppm improvement which nobody else can use.
The big advance is, that now the 9digit region has been reached.
Later, I will do noise mesurements over various NPLC numbers, and I will provide an Allan Deviation diagram.
Both will show that the 3458A now performs with about 5 times lower noise floor, and the Allan Deviation wil go down at a 20s time constant to about 4E-9 (2E-8 before) , which is  already this 9 digit performance.

Yes, 1281 and 8508A are identical and should both benefit from this improvement, as their A/D is similar to the 3458A (same working principle)
branadic already uttered his doubts about the PREMA 6048, because the A/D itself is supposed to need 20sec to produce 8 digits. This is as well the lowest possible aperture time for 8 digits, unfortunately.

This was not yet the root cause for the 500nV .. 1µV StD noise floor of my 1271, I have to investigate on that further.

You can do me a big favor,  and simply send me your StD figures for 7 and 8 Dig slow and fast modes.
You only need to take about 10 samples each, and calulate average and Standard Deviation , or you just send me the measurement data, please.

THX Frank