Ultra Precision Reference LTZ1000

[MisterDiodes]

Dr. Frank,
Do you have a way to test your LTZ's against something other than one 3458a?  I see that you've produced data, but normally we would never just use another single LTZ to measure against (aka 3458a).  For instance if you measure an LTZ circuit with just a single 3458a, in essence you're comparing only the drift of one LTZ die to other LTZ dies, using basically the same circuit for all Vref's.  The drift trend direction will tend to be the same, and you can possibly convince yourself into much better results if you're -only- looking at inter-comparing LTZ die.

At least you know you don't have a completely mucked up circuit module, but it is still wise to use other test methods if possible.  That's hard to do with an LTZ since that's already at close to the top of the food chain.

It would be interesting to see how the results compare to more than one reference technology to test against - for instance we use multiple cal'd 732's, KVD's, nullmeters & multiple 3458a's to measure a Vref accurately... You'll raise confidence and lower uncertainty of any LTZ  absolute value measure /  long term drift assessment that way - which is the whole point if you're down in the low ppm range.

Just another example: more tools ===>  happier Volt-Nut!!

[MisterDiodes]

Also, be careful using 3458a automatic ratio mode at low PPM - that can introduce more uncertainty.  You can still do ratio measures, but for lowest ppm uncertainty you'll generally take manual measures and calculate the ratio from those results (Or using a low thermal scanner switch setup, etc.)- Better yet if you can do a manual transfer measurements from a cal'd 732 (for instance) for best result if you're after low traceable uncertainty.  Explained here:

http://literature.cdn.keysight.com/litweb/pdf/5992-1058EN.pdf

Getting true a absolute value measure at low traceable uncertainty isn't trivial on a high performance LTZ - if that's what you're after.

[zhtoor]

Dr. Frank,
Do you have a way to test your LTZ's against something other than one 3458a?  I see that you've produced data, but normally we would never just use another single LTZ to measure against (aka 3458a).  For instance if you measure an LTZ circuit with just a single 3458a, in essence you're comparing only the drift of one LTZ die to other LTZ dies, using basically the same circuit for all Vref's.  The drift trend direction will tend to be the same, and you can possibly convince yourself into much better results if you're -only- looking at inter-comparing LTZ die.

At least you know you don't have a completely mucked up circuit module, but it is still wise to use other test methods if possible.  That's hard to do with an LTZ since that's already at close to the top of the food chain.

It would be interesting to see how the results compare to more than one reference technology to test against - for instance we use multiple cal'd 732's, KVD's, nullmeters & multiple 3458a's to measure a Vref accurately... You'll raise confidence and lower uncertainty of any LTZ  absolute value measure /  long term drift assessment that way - which is the whole point if you're down in the low ppm range.

Just another example: more tools ===>  happier Volt-Nut!!

hello,

short of testing against a JJA, would a set of weston cells make a viable choice to test LTZ1000's against?

regards.

[TiN]

There was a question about overdriving LTZ with high current. Well, I got some practical data, using jumper chip as a specimen, since I don't care much about it's long-term stability.
Graph data:

LTZ unit with "default" resistor and current.
Same LTZ with 60 ohm current resistor. - that's about 10 mA zener current.
Same LTZ with 40 ohm current resistor - that's about 15 mA zener current. Output jumped +855 ppm as result.
Difference in output between 60 ohm and 40 ohm current resistors.

[Edwin G. Pettis]

Zhtoor....

The Weston cell has long passed into history as a voltage standard, while you might get a Weston Cell calibrated at NIST to .04 PPM, that comes with a long list of limits, that is only valid at the time of measurement, any movement of the cell, even gently will invalidate that measurement and increase the uncertainty, not to mention the cost is in the thousands of dollars, they are very fragile and have significant temperature coefficients, you would get a better measurement comparing two LTZs against each other.  However, for a valid and accurate reading, they must be compared to a certified 732A (preferably two or more), otherwise you are just comparing drift rates between the LTZs.  The 3458A is a very good transfer standard, used to compare a LTZ against a 732A you can get sub PPM uncertainty if done correctly but the 3458A by itself is not a standard.

A functioning Weston Cell might give you a reference that is possibly good to perhaps 10 PPM or 20 PPM with a uncertainty of a few PPM if handled very carefully, that would have to be done on site to minimize uncertainty, you could probably get away with using a very good 731B in this case, an old one with a very good drift history.  You can't draw any significant current from it, you don't want to leave it connected to a meter for any longer than necessary and you want the temperature to be reasonably steady.  Way more trouble than a LTZ Vref, even one that isn't performing top notch.

[zhtoor]

Zhtoor....

The Weston cell has long passed into history as a voltage standard, while you might get a Weston Cell calibrated at NIST to .04 PPM, that comes with a long list of limits, that is only valid at the time of measurement, any movement of the cell, even gently will invalidate that measurement and increase the uncertainty, not to mention the cost is in the thousands of dollars, they are very fragile and have significant temperature coefficients, you would get a better measurement comparing two LTZs against each other.  However, for a valid and accurate reading, they must be compared to a certified 732A (preferably two or more), otherwise you are just comparing drift rates between the LTZs.  The 3458A is a very good transfer standard, used to compare a LTZ against a 732A you can get sub PPM uncertainty if done correctly but the 3458A by itself is not a standard.

A functioning Weston Cell might give you a reference that is possibly good to perhaps 10 PPM or 20 PPM with a uncertainty of a few PPM if handled very carefully, that would have to be done on site to minimize uncertainty, you could probably get away with using a very good 731B in this case, an old one with a very good drift history.  You can't draw any significant current from it, you don't want to leave it connected to a meter for any longer than necessary and you want the temperature to be reasonably steady.  Way more trouble than a LTZ Vref, even one that isn't performing top notch.

thanks and almost agreed.

i was talking about characterizing the long term *drift* performance of LTZ1000's by comparing to other *types* of standards,
like LTFLU based, SZA263 based, 1N829A based, JJA based or *maybe* Weston Cell based, assuming different technologies drift differently.
inadequacies of Weston Cells are abundant, but, noise performance and drift is not one of them (in a tightly controlled temperature).

best regards.

[Dr. Frank]

Dr. Frank,
Do you have a way to test your LTZ's against something other than one 3458a?  I see that you've produced data, but normally we would never just use another single LTZ to measure against (aka 3458a).  For instance if you measure an LTZ circuit with just a single 3458a, in essence you're comparing only the drift of one LTZ die to other LTZ dies, using basically the same circuit for all Vref's.  The drift trend direction will tend to be the same, and you can possibly convince yourself into much better results if you're -only- looking at inter-comparing LTZ die.

At least you know you don't have a completely mucked up circuit module, but it is still wise to use other test methods if possible.  That's hard to do with an LTZ since that's already at close to the top of the food chain.

It would be interesting to see how the results compare to more than one reference technology to test against - for instance we use multiple cal'd 732's, KVD's, nullmeters & multiple 3458a's to measure a Vref accurately... You'll raise confidence and lower uncertainty of any LTZ  absolute value measure /  long term drift assessment that way - which is the whole point if you're down in the low ppm range.

Just another example: more tools ===>  happier Volt-Nut!!

Hello Mr. Diodes,
your first question and remark is legitimate.
Yes, over the last years, I use the 3458A as a comparator and its LTZ1000A @ 65°C for the baseline reference, plus a Fluke 5442A, plus these two prototype LTZ1000 references.
The LTZs have been running for 12 years now, whereas the 3458A and the 5442A were only switched on when used.
So, the LTZs should have the typical -0.8ppm / year drift, the 3458A a fraction of that drift, down to zero drift, as the references typically do not drift when unpowered. Same goes for the 5442A, which should typically  have a positive drift over time.

Starting in 2009, I compared these 4 references regularly, and their relative drifts were as expected, i.e. the 5442A upwards, both LTZs downwards, relative to the 3458A.
The total relative drift per year between the 5442A and both LTZs was less than 1ppm, which was my stability goal.

That's the poor mans way of maintaining Volt, as I don't want to calibrate the standards by a metrology lab, or by usage of commercial instruments, like a 732A.
Btw.: The 100 year-drift of Le Grand K, the kg prototype, was identified by the very same principle.


I doubt, that a KVD or a Null Detector give any advantage over the 3458A as a comparator..

During that time, I had the chance to 'zero' the absolute Volt in my lab, at first when I got the 5442A, then by a comparison to a recently calibrated 8508A, and latest, when I got two brand new 34465A from KS, which agreed within 1ppm to my baseline.

That's all no regular calibration, that's crystal clear.
But for an amateur grade lab, there is high  probability, that these artefact calibration points, together with the frequent relative drift measurements, provide good enough accuracy at about a few ppm.
I don't want to say uncertainty, as this would require or imply regular calibration methods by a proven chain  from official metrology labs.

And I also want to emphasize, that I do not want to dig into the tenths of ppm region, where all these nice officially accepted standards rule.

The addition of up to five new LTZ1000 may improve the drift statistics, although they will also have a probable negative drift.. so SZA263 based references are lacking.

Anyhow, my current group of 4 references, with their underlying history will be sufficient to measure the drift behaviour of the new references, and that's the most important feature.

Now that I have some more compact LTZs, whith proven low T.C., I can send them to a friendly volt-nuts companion, who might punch another nail in my baseline.

Frank

PS: I also added a Hamon divider for 100:1 and 10:1 transfers (DIY), a 5450A, and recently a 845AR to my lab, and these redundant instruments  increased my confidence and insight a lot, over the bare uncertainty of the 3458A.

[MK]

There was a question about overdriving LTZ with high current. Well, I got some practical data, using jumper chip as a specimen, since I don't care much about it's long-term stability.
Graph data:

LTZ unit with "default" resistor and current.
Same LTZ with 60 ohm current resistor. - that's about 10 mA zener current.
Same LTZ with 40 ohm current resistor - that's about 15 mA zener current. Output jumped +855 ppm as result.
Difference in output between 60 ohm and 40 ohm current resistors.

Hmmm, it is a shame that the jumps did not stop, but it definitely seemed at first glance to be improved, at 15mA your jumpy specimin seems to have a much lower tempco, that stands out to me. Some stuff I have read is that at low currents a zener will jump like that, so it is just a mild die fault then? any experts know something that they are allowed to say? no NDA's in the way that is.

[martinr33]

Agilent, in Service Note 3458A-18A, states that unpowered LTZ1000 references can lose their ageing characteristic if stored unpowered, leading to an error of up to 15 ppm. Running the meter for 6 weeks apparently solves this problem. They don't say how long the unpowered period is. Dr. Frank's evidence would suggest that annual checking is enough. Or maybe, this problem is limited to very few units.
Plus, the 3458a operating temperature is higher, which raises the question of how a higher operating temp changes things in this regard.

Dr. Frank's numbers suggest that he, at least, does not have this problem. 15ppm would show up in that simple comparison to a new meter.




The internal reference assembly used in the 3458A are aged (stabilized under a powered-up state) and
monitored for appropriate DCV time drift for use in the 3458A. Initially, virtually all references drift
at rates too high for use in the 3458A. However, the rate of drift decreases with time and the future
drift performance of the reference assembly can be accurately predicted with a high level of certainty.
If the reference assemblies are powered down for an extended period of time (such as being stored in a
stock bin) the references revert back toward their “pre-aged state”. As a result the initial drift rate for
the DCV reference may be high enough that the DCV function may be “out of specification” prior to
the first calibration/adjustment interval (90 days or 1 year).
Some of our references used in recent production have been stored in a stock bin long enough to
exhibit this drift problem. The error associated with this long-term drift issue will cause a gain error
for all voltage, resistance and current measurements. This error is expected to be less than 15 ppm of
the reading.
Operation of the 3458A for six weeks is sufficient for these references to stabilize under a powered-up
state. Adjustment/calibration of the instrument after this initial six week operating period will
eliminate this issue and the reference will be sufficiently stable to remain within specification until the
next regular scheduled adjustment/calibration event.

[Dr. Frank]

Agilent, in Service Note 3458A-18A, states that unpowered LTZ1000 references can lose their ageing characteristic if stored unpowered, leading to an error of up to 15 ppm. Running the meter for 6 weeks apparently solves this problem. They don't say how long the unpowered period is. Dr. Frank's evidence would suggest that annual checking is enough. Or maybe, this problem is limited to very few units.
Plus, the 3458a operating temperature is higher, which raises the question of how a higher operating temp changes things in this regard.

Dr. Frank's numbers suggest that he, at least, does not have this problem. 15ppm would show up in that simple comparison to a new meter.




The internal reference assembly used in the 3458A are aged (stabilized under a powered-up state) and
monitored for appropriate DCV time drift for use in the 3458A. Initially, virtually all references drift
at rates too high for use in the 3458A. However, the rate of drift decreases with time and the future
drift performance of the reference assembly can be accurately predicted with a high level of certainty.
If the reference assemblies are powered down for an extended period of time (such as being stored in a
stock bin) the references revert back toward their “pre-aged state”. As a result the initial drift rate for
the DCV reference may be high enough that the DCV function may be “out of specification” prior to
the first calibration/adjustment interval (90 days or 1 year).
Some of our references used in recent production have been stored in a stock bin long enough to
exhibit this drift problem. The error associated with this long-term drift issue will cause a gain error
for all voltage, resistance and current measurements. This error is expected to be less than 15 ppm of
the reading.
Operation of the 3458A for six weeks is sufficient for these references to stabilize under a powered-up
state. Adjustment/calibration of the instrument after this initial six week operating period will
eliminate this issue and the reference will be sufficiently stable to remain within specification until the
next regular scheduled adjustment/calibration event.

This applies to the hp LTZ1000A circuit, which runs at about 90°C.
This implies a much higher drift rate, but also a hysteresis, or creep effect.
What I have seen on my LTZ1000 references, that both chips do not show noteworthy hysteresis, if the oven temperature is not that high, so 65°C as intended will give a small hysteresis, and 45°C, as in Pickerings design of the Datron 4910 or the 7001 references, will give no hysteresis.

So, that's another reason, why I have pimped my 3458A reference to about 65°C.

Frank

[TiN]

I tested for this once too, did not see the effect, however I run both of my meters pimped too (100K VPG resistor over stock 15K one).

[martinr33]

I've moved one to the lower temperature as well. I figure if it gets hot enough to be a problem, I won't be using the meter that day!

I guess these were designed before the long-term characteristics of the LTZ1000 were fully understood. The design does seem to run unnecessarily hot, and I wonder if that gets in the way of long-term stability.


An interesting observation about these meters is that there seems to be no airflow over much other than the power supplies. There's nothing to clean off of the boards. The shielding seems designed that way. So I am thinking that the inguard boards are sufficiently low power that there's minimal self-heating going on. Maybe 10 - 12 Celsius, judging by the internal temperature measurement. Even running in a rack, the enclosure would have to be really hot to cause problem with the reference. In practice, I would be worried about the power supplies if the meter got hot enough to cause a problem.

[TiN]

As it was covered many times, HP deliberately designed meter to fit +50C operation temperature range, for applications like production, where meters sit packed in the hot rack.

[martinr33]

Max operating temperature according to the apc is 55C (storage up to 75C). With a 15 degree internal rise, the reference would have to be certain to operate at 70 celsius. 90 celsius allows for an extra 20 degrees of guardband, which would only make sense if the operating temperature was hard to predict or manage. But HP certainly knew the characteristics of these parts.

Plus, at that time, nobody would have known about the issues with operating the LTZ1000 at the raised temperature.

Also, operating at 55C (which is really hot, rack or not), the lifetime of the capacitors subject to high ripple currents would have been awful.

Bottom line is, we can safely drop the operating temperature of the LTZ1000 to 65C, and the meter may still be OK at 55C operating temperature.

[Dr. Frank]

Max operating temperature according to the apc is 55C (storage up to 75C). With a 15 degree internal rise, the reference would have to be certain to operate at 70 celsius. 90 celsius allows for an extra 20 degrees of guardband, which would only make sense if the operating temperature was hard to predict or manage. But HP certainly knew the characteristics of these parts.

Plus, at that time, nobody would have known about the issues with operating the LTZ1000 at the raised temperature.

Also, operating at 55C (which is really hot, rack or not), the lifetime of the capacitors subject to high ripple currents would have been awful.

Bottom line is, we can safely drop the operating temperature of the LTZ1000 to 65C, and the meter may still be OK at 55C operating temperature.

Your statements are utterly wrong and dangerous!   

You should have read the correct facts from experienced users beforehand, as you have to take into account more sources of temperature rise.

The minimum temperature rises are:
+13°C between room and internal temperature, due to instrument self heating
+10°C self heating of the A version
+5°C heater regulation margin

Therefore, this modified 65°C oven temperature limits operation to 37°C ambient!!

Then, you have to take into account additionally:
+5.. +10°C contamination of fan
+5.. +10°C usage inside a 19" rack
10°C production variation of LTZ1000 regarding oven set temperature.

So, in total, you might easily have 40°C temperature difference between ambient and oven temperature, that's why hp initially set that to about 95°C.

At last, I recommend this modification to 65°C oven temperature only by carefully monitoring the internal temperature, which must stay below 45°C.

I further recommend a max. ambient operating temperature of 35°C, operation on desktop, not in a rack, and frequent cleaning of the fan.


Frank

[tszaboo]

Not to mention, the internal temperature is avaliable as a GPIB command from the 3458A.
So if you want to use it as a high end, accurate, desktop meter maybe used for calibration, 1000 NPLC, you can always just request the internal temperature.

The script I've used for calibration did a (3458A) self calibration every time the internal temperature changed by 1 degrees.

[Kleinstein]

For best accuracy one should have the filter clean on the 3458 anyway. With a clogged filter the airflow is reduced and thus different temperature gradients are established. For this reason a temperature regulated fan is not a good idea in a design where the airflow is directly across sensitive parts. It could be if the sensitive part would be fully protected from the airflow, and airflow dependent temperature gradients.

Reducing the temperature for the reference slows down the drift and settling of the reference. So when cold it also takes longer for the reference to settle. So after something like 2 years the higher temperature reference might be more stable, as it can be more settled. A lower temperature is not always an advantage.

[branadic]

A new LTZ design is in town, this time made by Analog Devices / Linear Technology themself for AD5791 and with all the things you can think of such as slots, BMF resistors, ...



Source

-branadic-

[floobydust]

Didn't the previous eval board have the LTZ on it: https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg828843/#msg828843

[MisterDiodes]

Yes the AD5791 demo board has always had a place for LTZ1000 to be installed, for many years.  If you look at the datasheet they never really mention the VRef to use with that chip, nor do they mention the fact that the "low noise" figures quoted aren't for low freq / DC - and watch how the spec'd noise units change at low frequency.  Lots of red flags.  At the time the chip came out AD had no Vref to even come close to driving that Dac to full potential, so they quietly just put the competitors (LT) Vref part on the demo board.  Some of the boards have the LTZ layout upside down, I guess people sometimes don't catch the TOP SIDE vs BOTTOM SIDE pinout view on the LTZ datatsheet.  It happens.

The demo board I have in front of me also has a non-descript connector for "Vref input" with no description of how important that is if you want the full DAC specs...like any old voltage source will do.  A 7805 should work fine, right?

As with all diffused IC resistors the DAC noise is pretty high - if you're trying to use this chip to make some sort of an adjustable -DC- voltage source, I wouldn't do that if low noise is important to you.  The chip is very sensitive to board stress also.  It didn't impress anyone very much at the lab.
 

[Echo88]

Some guy on http://bbs.38hot.net/ made working kits of the AN86 suggested by Jim Williams: http://bbs.38hot.net/thread-29599-1-1.html But it seems he doesnt produce kits anymore. Maybe someone here has aquired such a kit or knows about the thread?

[martinr33]

You raise a really important point: I need to put a label on the  unit that indicates that it is modified for low temperature operation.

And yes, I missed the extra 10 degrees of latitude recommended for the A version. I may end up testing that next summer, as temperatures rise. I'll be watching to see if my reference appears to drift as we cross the 37C threshhold.

Something else: I need to dig into  the unit again, but it looks like the airflow design is mostly to keep the power supplies cool. There seems to be little flow over the other boards. In photos online, the units do seem to be very clean inside, and even the filters don't appear that dirty.

And finally - the deeper air filter on the older units does seem to let them run a lot quieter than the new filters, which are right on top of the fan. 

[hwj-d]

... and this one:

[mimmus78]

In last video they measured noise in the 1uV range with 3458a? Isn't the meter contribution the great part of the noise?

[TiN]

3458A was set to fast NPLC, so it's hard to say.

Meawhile, jumper LTZ driven with 15mA still alive and kicking. With some heavy averaging it can even seem to be almost normal:



There are step temperature excursions, stable at levels to 0.005C with points +21C, +50C, +30C and +24C.

CSV-data in case somebody wants.