Ultra Precision Reference LTZ1000

[Martinn]

So I calculate maximum 50 hours operating time for a fresh set of batteries.

As I find adding a battery a major PITA - what do you think are the effects of switching off the reference (hysteresis)? Mostly I'd have it running (at home), probably biggest influence would be bringing the unit to my local calibration lab (I guess I can tell them to run the unit overnight before measuring).
Of course the battery option nicely solves the isolation problem.

Maybe it's not too relevant for me, I only have a 6.5 digit DMM, although performance is quite decent: I measured a 7 V reference (ADR1399 eval board), averaging gave 7 digits, standard deviation of 1.3 uV. Specified accuracy is much lower of course, but for comparisons or ratio measurements I guess it is not too bad.

[Kleinstein]

The LTZ1000 does show some hysteresist in heating cooling cycles.  So a short (e.g. 1 minute) power down can have an effect on the voltage and it may take weeks to recover from this. The details depend on the set temperature.  So battery supply makes some sense. A reasonable solution can be 2 inputs for power linked with diodes. So one can change from one power source to another without interuption or could have a slightly lower voltage battry back as a backup in case the mains supply is interrupted.

[Martinn]

What amount of change can one expect as hysteresis? Is this in the sub-ppm region or more?

[Dr. Frank]

Usually, you observe sub-ppm or no change. The LTZ1000 may show bigger effect than the LTZ1000A. I always shut off my 3458A, when it's not in use, and observe absolutely zero hysteresis, after the usual 2h warm up. I lowered it's oven temp to about 65..70°C, to reduce it's drift
Both types are very sensitive to storage temperatures below 10°C, and must not be stored < -10°C, which creates a lot of hysteresis to about +1..5ppm, in latter case this might disable the hysteresis removal process by J.R. Pickering, in his Fluke 7000. See description in manual.
I have no problem with all of my LTZ1000  references , all running at 45..55°C, when I transport them "cold" in summer time. There's virtually no hysteresis at all.
I have not heard of any such problems as well from the U.S. Cal Club.
Frank

[dietert1]

Of course it makes a difference whether you run the reference 20 K or 60K above ambient temperature.
During my current study of a LTZ array i am using an external oven in the 30 to 45 °C range and after one or two temperature sweeps up and down all references come back within fractions of a ppm. Then they do not exhibit hysteresis anymore.
Hysteresis may also stem from metal foil resistors used in those reference circuits.

Regards, Dieter

[Martinn]

Sounds fine... I plan on running an LTZ1000A with 12.5k temperature setpoint in my office, no arctic travels (and the cal lab might be in extended walking distance), so I guess I can live without battery backup.
Makes me wonder if I could measure a 1 ppm step. DMM6500 has 10 uV least digit (1.4 ppm) and 1.3 uV std of the averaged readout (0.2 ppm).
For that I probably can't even look unevenly at the connectors, otherwise they'd introduce thermal voltages far above that.
Has anyone considered wearing thick insulated winter gloves when handling connectors to reduce their heating?

Hysteresis may also stem from metal foil resistors used in those reference circuits.

So let's hope Edwin can still fabricate the precision wirewound resistors necessary...

[Dr. Frank]

1. use a differential method, resolving down to 10nV.
You'll need at least 3 references, anyhow
2. use CuTe jacks and appropriate connectors like Au coated or pure copper lugs or 4mm plugs of similar metal composition.
3. induced e.m.f. voltages disappear usually after a few tens of seconds.
4. Always use a set of 10..16 readings, with statistics, so you can judge from the StD value, if you measured in a stable manner. 100..300nV StD @ 7.15V level is fine. That might not be achievable with an ordinary 6.5 digits DMM
5. always make a reversal measurement, to remove e.m.f., and determine the e.m.f. value
6. you might also use 12k/1k for 45..55°C, like in the venerable Fluke 7000
7. I made very good experience with standard values of PWW resistors from G.R. or BMF from AE. both came from rhopoint in UK
I use 1k, 12k, 2x 70k and 120 because of the reduced oven temperature,  compared to the bad hp design

[Martinn]

CuTe jacks and cables: I was planning to use Pomona 3770 binding posts and also buy some low emf cables from Mouser (any hints...?)

3 references: Andreas already mentioned that. Makes sense, but current LTZ1000A price at mouser is near $100 and custom resistors on top... will in the end be as expensive as my 6.5 digit bench DMM.

Statistics: Definitely. One thing I really like with the DMM6500 is the graphical display - you immediately see if there's drift.
100 nV StD: Sorry, I don't (yet) have the 8.5 digit entry ticket for the inner voltnut circle... 1 uV is the best I can get (5 NPLC, 10x filter)!

12k/1k: I've read a couple of times this might get a bit tight at elevated room temperatures, that's why I assumed 12.5k.
Assuming a metal box (probably aluminium, although steel would be better for magnetic shielding) does it help to add some foam isolation inside over the entire PCB? I have seen some builds like that. This would probably reduce the margin for the LTZ heater (if the entire PCB would heat up inside its isolation)? Or just have some draft shield over the LTZ1000A within a otherwise non-insulated box?

Rhopoint UK: Thanks a lot - excellent hint! GR 8G16 seem to be actually available in many values (not 12.5k however)! No confirmation yet from Edwin, he was hinting at some wire sizes running out.

[Dr. Frank]

your DMM will resolve 100nV in differential Mode, in it's 1V or 100mV at least
I use double shield, I.e. a tuner box with foam  and an outer aluminum box. 
Andreas uses a similar shield.
So I can use case ground and Guard. See my assembly of  the 4fold references.
I use old mc cables, and assembled shielded PTFE with either lugs or 4mm banana from mc
shielded PTFE gave a noise improvement especially for OHM measurements

[antintedo]

What amount of change can one expect as hysteresis? Is this in the sub-ppm region or more?

Long time ago I did a small experiment to determine the effects of shipping the reference unpowered in colder seasons, maybe it will be helpful. The effect seems to be <0.5ppm if the reference never saw low temperatures, slightly more if it did. Degaussing-like temperature swings seem to restore the original behavior after exposure to cold. Simply waiting was very slow, power cycles with cooling to ambient temp helped only a little. I tested two parts from the same batch with roughly similar behavior, here's one chart I found:



Lowering the oven temperature is a good idea but I haven't tested it. An extreme version of this idea would be putting the board inside a slightly cooled peltier oven and setting the LTZ to 25*C so the LTZ die never experiences significant swings in either direction. Perhaps it could allow shipping in near freezing temperatures, assuming the problem lies with temperature swings alone and not absolute temperatures as well.

[iMo]

I've recently added 20V Li-Ion (external, 5x 18650 old recycled batteries) to myADR1001, it lasts 2 days powering the box (37mA), wired via a diode to the voltage regulator's input of my built in linear psu (thus the battery and/or AC could be on). It is advisable not only to avoid power-off, but also to avoid AC leakage via the transformer during a measurement (powered from the batteries during the measurement). Your meter may resolve sub uV values after some math, but with its LM399 (afaik) inside you will see the LM399 fluctuations (a random walk/jumps within 5-10uVpp).

[dietert1]

As a heater oven is less circuitry than the Peltier oven, i was aiming for 37 °C, assuming it would be safe.

There is a fan inside the incubator i am using as temperature chamber for my array study and the six LTZ1000 references started behaving like an ants circus once the TC fine tuning approached 0.1 ppm/K. So i put the array into a little plastic container inside the temperature chamber to protect it from air draft. Then it's temperature happened to rise by 4 °C. I know this from the residual nonlinear TC.

So, once making an external oven, using low temperatures is possible and safe with a peltier heater/cooler. And i will measure actual chip temperatures of the references. The built-in second transistor can be a G=5 amplifier, so the output voltage will be about 2.5 V with -10 mV/K. If i digitize this with a 12 bit ADC with 3.3 V Vref, temperature resolution approaches 0.1 °C.

Regards, Dieter

[Martinn]

I've recently added 20V Li-Ion (external, 5x 18650 old recycled batteries)

I also considered that. However unless you have really well matched quality cells, you'd need an active balancer. Also Li-Ion does not like float charge, so some extra charge management is necessary.
Or slap in a bunch of alkaline AA? At least you can measure their state of charge easily. Or NiMH LSD (like Andreas) - but charging them propery is far from trivial and I guess extended float charging will damage the cells as well.

Main benefit I see is perfect galvanic isolation, no mains coupling problems.

ADR1001

Yes, this would be the other attractive option! Just wait another few months (end 23?), get a ADR1001 and be done with it. No resistor sourcing troubles...

your DMM will resolve 100nV in differential Mode, in it's 1V or 100mV at least
I use double shield, I.e. a tuner box with foam  and an outer aluminum box. 
Andreas uses a similar shield.
So I can use case ground and Guard. See my assembly of  the 4fold references.

Quad: I remember TiN's design. Could you post a link to yours? I wonder how you implemented the guard.
Just called the local calibration lab. They can measure (3458A) with an uncertainty of 10 ppm + 1.4 uV, about half an hour of labor. Cost sounds acceptable, but for multiple LTZs (and yearly) too much. So I'd calibrate probably just one and then do a comparison. The national lab (METAS) can do (much) better, for probably much higher cost.

[Dr. Frank]

instead of 12k5/1k, use 10k/800 

[Dr. Frank]

Quad: I remember TiN's design. Could you post a link to yours? I wonder how you implemented the guard.
Just called the local calibration lab. They can measure (3458A) with an uncertainty of 10 ppm + 1.4 uV, about half an hour of labor. Cost sounds acceptable, but for multiple LTZs (and yearly) too much. So I'd calibrate probably just one and then do a comparison. The national lab (METAS) can do (much) better, for probably much higher cost.

Let me GOGLE it ("Guard") for you: https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg3886166/#msg3886166

With a stable ensemble of 3 references, you would only need to send one of these for calibration, best < 1ppm uncertainty.
On return, you compare again, if the travelling standard has changed on its tour, and then you could calibrate the others on your own.
Hint: Adrian from AB Precision makes reasonable prices for volt-nuts, but w/o official certificate. 10ppm is quite mediocre, Adrian can do better.
Another hint: A 3458A is NOT a reference standard! Therefore such a "calibration" I cannot recommend.
The 3458A can be used as a transfer standard < 1ppm against another "true" reference.

Frank

[Dr. Frank]

@ Martinn, another hint: purchase the LTZ1000, i.e. the non-A version.
It's cheaper, and you can use 45..55°C oven temperature, with foam insulation, even up to 35°C and more room temperature w/o a problem.
Please search for yourself, this time  https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg3263778/#msg3263778.. in this thread I have shown T.C. measurements on one of my LTZ1000A references, using 12k/1k as well, and demonstrated that these failed at an internal temperature of > 39°C. The thermal isolation provides another ~7K temperature difference, therefore a bit more than 30°C r.t. is feasible.
Anyhow, precision measurements should only be made between 18 .. 28°C

Frank

[Martinn]

Please search for yourself, this time

I expected this type of comment - it was inevitable...
Actually all questions one might ask on LTZ1000 have probably been answered several times in the past 3200 posts. Unfortunately it is near impossible to find the answers, Google does not help (unless you know some magical key word) and ChatGPT unfortunately has no clue of EE.
So thanks a lot for answering anyway!

Regarding the non-A version: Would not the better thermal coupling of the non-A lead to longer stabilization times and reduce the immunity to external temperature deviations? Or is the temperature control (an I controller as I understand it?) good enough so it does not matter either way? Combined with your foam isolation, would that not result in a much longer stabilization time?

Guard: Interesting, first time I see this. It seems that precision DMMs (my 6.5 digit does not have a guard terminal) interpret "Guard" differently than in high impedance (e.g. pH probe) measurements: With those, you drive exactly one signal line guard to the level you measure it to be. In DMMs the guard seems to enclose the whole input section, not just one signal. That probably makes sense as they have an excellent isolation to PE (earth) and their guard can be driven to some convenient potential (e.g. LO input).

I have to say I am a bit demotivated by the lowish 10 ppm uncertainty I could get from my local cal lab. After all, setting up several LTZ1000 boards is quite a (expensive) effort. On the other hand, simply switching on my DMM6500 (still in cal) will give me a 32 ppm uncertainty reading a 7.2 V source.

- Martin

[Andreas]

Hello,

Or is the temperature control (an I controller as I understand it?) good enough so it does not matter either way?

The I part of the controller is low compared to the thermal time constants.
And yes the "400K" resistor compensates the tempco for the non-A version.

I have to say I am a bit demotivated by the lowish 10 ppm uncertainty

Perhaps you should start with some LM399 references which are cheaper and after some run in time the ageing drift is comparable.
(you should only not tilt them since they are not fully symmetrically like LTZ1000)

with best regards

Andreas

[Martinn]

Hello Andreas,

Perhaps you should start with some LM399 references which are cheaper and after some run in time the ageing drift is comparable.
(you should only not tilt them since they are not fully symmetrically like LTZ1000)

Well, I already have a DMM6500 which has a LM399 built in, possibly selected/pre-aged (?), so I don't see much benefit in adding another one. A somehow cool idea using the LTZ1000 would be to have a few-ppm accurate reference around, but for now I don't see an affordable way of getting a sufficiently accurate reference measurement. I will certainly not spend a few hundred Fr. for a Josephson opinion from the national lab.

Tilt sensitivity: This would apply to the ADR1399 also? I have an eval board of the LCC version, have to try this.
BTW I ran this ADR1399 module also on a DMM7500 I have limited access to, I saw some discrete steps (forgot the amplitude, but significantly above the standard deviation of both the DMM7500 and my DMM6500) I don't see on my 6500. I have seen comments that the LM399 can do this, however it's strange I don't see them measuring with my DMM6500. Maybe it was just some warmup hiccups of the 7500 as it sat unpowered in a shelf before.

- Martin

[Andreas]

for tilting of ADR1399LS8 see here my measurements:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg4598620/#msg4598620

of course the results depends on thermal isolation. (will get worse with less isolation).

with best regards

Andreas

[Kleinstein]

Without proper meter or other system to compare there is little value in having just 1 LTZ1000 reference. At least one could not check the aging of the LTZ1000 reference.
One could have luck with 2 LTZ refrence: one well aged and than look at the difference for a new one or with heating 1 of the 2 circuits.
It could really make sense to gain experience with LM399 or ADR1399 reference first.

The better isolation of the A version could effect the quality of the temperature regulation with a little more loop gain. I think the more relevant point is reducing stress the the chip from a kind of soft mounting and also less variations in the heater power.

The resistive heater has a square law and thus a loop gain going up with more heater power. AFAIK the residual TC of the non A version before the compensation with R9 is more linear and not getting clearly better with higher heater power or much worse when at low power. This indicates that the residual TC is not from insuficient loop gain in the temperature control loop, but more from thermal gradients inside the chip caused by the heater power / sensor not at the ideal position. The smaller thermal resistance means more variation in the heater power, though this difference alone is only about 2.5-5 fold  (the data are abit confusing showing a 5 fold ratio in the termal resistance data in the table, but closer to 2.5 fold in the power needed in the heating curves.

[Dr. Frank]

Please search for yourself, this time

I expected this type of comment - it was inevitable...

Sorry, den Spruch konnte ich mir diesmal nicht verkneifen. 

Regarding the non-A version: Would not the better thermal coupling of the non-A lead to longer stabilization times and reduce the immunity to external temperature deviations? Or is the temperature control (an I controller as I understand it?) good enough so it does not matter either way? Combined with your foam isolation, would that not result in a much longer stabilization time?

The LTZ1000A stabilizes its oven temperature within about 3 sec, the LTZ1000 requires about 30sec. I provided an overview table for all of my different references in this thread.. I don't repeat that sentence again.. 
Thermal stabilization of the complete circuit, i.e. including the 5 precision resistors is virtually identical between A and and non-A version, i.e. within 1h.

 

Guard: Interesting, first time I see this. It seems that precision DMMs (my 6.5 digit does not have a guard terminal) interpret "Guard" differently than in high impedance (e.g. pH probe) measurements: With those, you drive exactly one signal line guard to the level you measure it to be. In DMMs the guard seems to enclose the whole input section, not just one signal. That probably makes sense as they have an excellent isolation to PE (earth) and their guard can be driven to some convenient potential (e.g. LO input).

Well, higher grade instruments have exactly the same guarding scheme, look how the HP3458A, Fluke 335, 720A, 752A are built.
It's not that complicated, and as it's a passive Guard in each case, this might not be as effective.
I see an improvement in noise immunity, anyhow.

I just upgraded my "cold" travelling reference LTZ #3, which has an LTZ1000 inside.
I tested it's warm-up behavior after a 56h off-time @ 21°C room temperature. It returns to its initial value to +/- 0.1 ppm within 1h, see first diagram.

In the longterm stability diagram, you can see two  hysteresis shifts due to low temperature storage (<0°C).
At the end, there was a temporary shift of + 0.3ppm directly after assembling LTZ#3 into its new outer case. After one week, its reference value returned to its recent one within 0.1ppm, as well after the 56h off time.

So it's ready for a comparison @ Adrian.

I have to say I am a bit demotivated by the lowish 10 ppm uncertainty I could get from my local cal lab. After all, setting up several LTZ1000 boards is quite a (expensive) effort. On the other hand, simply switching on my DMM6500 (still in cal) will give me a 32 ppm uncertainty reading a 7.2 V source.

- Martin

At first, build your LTZ reference. Then you might either ship it to Adrian for a volt-nuts level calibration, <1ppm "unofficial" uncertainty, or you might join our Cal-Club and make an exchange with either of us.  It's a rather simple procedure, I guess, although you're living in Switzerland. You need an anchor reference in your lab, to identify possible shifts during transport. That might be your DMM in first place. Another LM399 reference would be better.

[iMo]

..I have to say I am a bit demotivated by the lowish 10 ppm uncertainty I could get from my local cal lab. After all, setting up several LTZ1000 boards is quite a (expensive) effort. On the other hand, simply switching on my DMM6500 (still in cal) will give me a 32 ppm uncertainty reading a 7.2 V source.

You have to distinguish between a "metrological business" and a "hobby measurement". From my experience the concrete people in the labs there are kind and helpful people, sometimes they are voltnuts too.
99.999% of the routine boring measurements they do daily are for the clients who need their 5.5-6.5digits meters "calibrated" - where it means they need an "official" confirmation on paper their meter is within a spec (for legal purposes). Thus their uncertainties have the "legal" component built in (to be on the "safe side" in case of a dispute). And also they make $$ with that activity, indeed. When you talk to them (to a certain lab person there) as a hobbyist who does not require "legal paper" and all those boring calibration activities around a meter and just wants to see a voltage/frequency once a year/two they will certainly help you. Again - the people there are mostly nice and kind, from my experience.. 

[DH7DN]

"Hobby measurement" basically means everything and nothing at the same time.  If you ask a hobbyist "How much confidence do you have in your measurements?", the answers might be very different - depending on who you ask: "Everything is in spec", "Equipment recently calibrated against a well-known standard with an uncertainty of ...", "10.0000000 V as you can see". I think Dr. Frank's description is pretty much on point.

[iMo]

A "hobby measurement" of a single Vref source in a professional metrology lab could be made with a better uncertainty compared with the "lab's official uncertainty" - I am pretty confident about that..