HP 3458A ; A/D board spare part availibilty and cost

[MiDi]

My unit has datecodes from end '89, everything is original, only a couple of replacements were made - details in 3458A worklog thread.
The U180 is a hybrid consisting of a resistor array and switches, a die shot of U180 is somewhere in one article of 3458A on xdevs - Edit: added link to die shot from xdevs article.

It is not so easy and costly to determine drift of reference with low uncertainty, but soon  I want to post some data of the unit.

[dietert1]

For an old instrument even slow aging should have come to an end, except if it depends on operating temperature and the instrument was essentially off all those years. Did you think about trying a TEC pile on that ADC hybrid U180? I mean if the firmware of the 3458A does not support temperature and drift corrections, maybe you can do it by chip temperature control. If aging depends on chip temperature, one could try running that chip a little cooler.

Before trying mods on an expensive instrument i would decide upon a certain goal in terms of ppm/year or so.

Regards, Dieter

[MiDi]

Not shure what you mean by "aging should have come to an end", even best secondary standards are drifting after aging for dozens of years (e.g. SR104, 732x) 
If the drift rate is known, quite stable and somehow predictable you are able to include that into the calculation of measurement result and uncertainty - same applies for TC.

I doubt it is worth the effort in trying to compensate for drift and/or TC for U180 - imo ACAL already does this (and much more) satisfactorily.

Keep in mind, 3458A is not a secondary standard - the internal references are not designed for best TC nor drift, but it is one of the best ratio measurement instrument.

[Dr. Frank]

For an old instrument even slow aging should have come to an end, except if it depends on operating temperature and the instrument was essentially off all those years. Did you think about trying a TEC pile on that ADC hybrid U180? I mean if the firmware of the 3458A does not support temperature and drift corrections, maybe you can do it by chip temperature control. If aging depends on chip temperature, one could try running that chip a little cooler.

Before trying mods on an expensive instrument i would decide upon a certain goal in terms of ppm/year or so.

Regards, Dieter

Hello Dieter,
Midi already wrote it as it is: No DMM is intended and accepted as (secondary) standard.
The 3458A is used in metrology mainly as a ratio machine for transfers, due to its ultra linear A/D.

Its voltage reference is neither optimized for timely, nor for temperature drifts.

As it's got 90°C oven temperature, drift would be usually about -30ppm/yr, but the complete modules are monitored and selected for annual drift of 8ppm, 4ppm or until recently 2ppm/yr.

The standard reference will usually drift less, like 2ppm/yr., after one or two years, it's rumored.
But it will not approach zero.
All drift specifications are valid for continuous operation, so it's probable, that the drift is very low when not in use, and if no hysteresis effects occur.

But even real voltage references, like the 732A/B/C, or the M7000, running only at 45..55°C, mostly show a constant drift, after years of operation. The LTZ1000 drifts typically  -0.8ppm/yr., and might settle down to - 0.5 ..-0.2 ppm/yr.

hp did not even trim the module for zero T.C., so the, I think, 0.3ppm/K are quite mediocre.

With a little bit of effort, it could have easily be tuned to 1/10 of that value.

The U180 has additionally about 0.5ppm/K, which can easily be compensated by the ACAL procedure.

hp did not intend to design a real metrology instrument, that's obvious.

So a bit of tuning can be done, like decreasing the oven temperature, maybe compensating the T.C. of the LTZ circuit. But then the story ends, and you better get a real 10V standard, which you might use for comparison, or for frequent calibration of the DCV mode.

Frank     

[dietert1]

Yes, i understood that the 3458A has a firmware procedure "ACAL" that takes care of ADC temperature and time drift. Anyway, this thread is about the risk of loosing the ADC and apparently an observed drift of 0,03 ppm/day (about 10 ppm/year) of the ADC alone makes people nervous. That's why i am asking: At which temperature is that ADC hybrid running? Is the temperature sensor of the instrument inside that hybrid?

Side remark: Tektronix 2465 scopes have a notorious hybrid U800 failure. It also happened in our lab and after fixing the instrument and solving the underlying heat/mount problem i thought: What a stupid mistake in an otherwise wonderful design.

Regards, Dieter

[Kleinstein]

The U180 hybrid should have low power loss and should thus not run much hotter than the general board. The problem is not so much the normal drift rate (e.g. < 0.1 ppm/day). This is easy compensated by ACAL every day or so. If the instrument was not in use for some time there may be additional drift for some time, but this would settle after a few days / weeks.

The problem is a possible and not so rare defect in U180 that is first visible as a much higher drift (often > 1 ppm/day). Chances are after some time the ADC would not work at all. So compensating this higher drift rate is not helping. The drift is only the symptom of a slowly failing ADC that is easiest and early to detect. So one can get an early warning, before the ADC will finally fully fail. With the high drift rate I would not fully trust the ADC anymore, as there can be other problems like higher DNL.

[Dr. Frank]

Yes, i understood that the 3458A has a firmware procedure "ACAL" that takes care of ADC temperature and time drift. Anyway, this thread is about the risk of loosing the ADC and apparently an observed drift of 0,03 ppm/day (about 10 ppm/year) of the ADC alone makes people nervous. That's why i am asking: At which temperature is that ADC hybrid running? Is the temperature sensor of the instrument inside that hybrid?

..

Regards, Dieter

Please first have a look into the Service Note 18, the criterion by hp was  0.43ppm/day (not 0.03ppm/day).

There are two flaws in this,
1) This 0.43ppm/day criterion is too big, only safeguarding the daily specified timely drift of the LTZ1000. 
2) this CAL? 72 parameter is also temperature dependent, that is not mentioned. As the usual T.C.s of the LTZ and of the U180 itself are in the same ballpark or even exceed this drift rate at reasonable lab temperature stabilities, its hard to separate this timely drift effect of U180 from these temperature drift effects.
It can be done with a lot of statistics, including the interior temperature, TEMP?.
Unfortunately, this T-sensor sits on the analog board on top, whereas U180 is located upside down on the bottom part of the instrument.

I think this issue is: 

Frank

[e61_phil]

I don't understand why they consider the LTZ1000 drift. A drift of the LTZ1000 shouldn't influence CAL72.

[Dr. Frank]

I don't understand why they consider the LTZ1000 drift. A drift of the LTZ1000 shouldn't influence CAL72.

That's right.
hp at that time simply used the drift specification: 24h / 90d / 1yr. (caused by the LTZ1000A drift) as a criterion for the allowable drift of U180, as latter daily drift should not exceed that specification either.

0.43ppm/day is the exact drift rate of the LTZ, interpolated to 1 day, if an annual drift of about 8ppm/yr. is assumed, and a square root drift dependency is used.
Frank

[dietert1]

Whatever Agilent/Keysight write, Midi's records demonstrate the ADC aging isn't a random process but something else and very regular. And it does reach 10 ppm/year at a daily drift rate of 0.03 ppm in his case. Hard to believe nobody tried to find out what that process is (and how to stop it).

Regards, Dieter

[Villain]

Whatever Agilent/Keysight write, Midi's records demonstrate the ADC aging isn't a random process but something else and very regular. And it does reach 10 ppm/year at a daily drift rate of 0.03 ppm in his case. Hard to believe nobody tried to find out what that process is (and how to stop it).

Regards, Dieter

Why would anyone make the effort when the 3458A is not a primary standard anyway (and not intended to be).
If i want more accuracy for DC i hook up a Fluke 732 and do a differential measurement using the transfer accuracy of the 3458A. Thats a sub-ppm measurement right there, no need to go through the trouble trying to get the 3458A in that region by itself.


edit: Or don't use a 3458A at all, only 732, dividers and null detector.

[niner_007]

My 3458A came back from Keysight. They replaced A1 DC board, A3 ADC board, and A7 display board, the EMI filter, fan, and mechanical switches. This is very thoughtful actually, so I have a new display too This was all included in the initial repair service contract, amazing experience. I've sent it in a pretty shitty box, and it came back in this amazing box They tested and calibrated, and declared it "like new" factory state. This is great The cal data looks amazing. Now we'll see how it drifts. I'll take it apart at some point to see the replacement boards, but hesitant given the very nice cal sticker on the screws.

[Villain]

My 3458A came back from Keysight. They replaced A1 DC board, A3 ADC board, and A7 display board, the EMI filter, fan, and mechanical switches. This is very thoughtful actually, so I have a new display too This was all included in the initial repair service contract, amazing experience. I've sent it in a pretty shitty box, and it came back in this amazing box They tested and calibrated, and declared it "like new" factory state. This is great The cal data looks amazing. Now we'll see how it drifts. I'll take it apart at some point to see the replacement boards, but hesitant given the very nice cal sticker on the screws.

Which Keysight Lab did you send your 3458A to?

[dietert1]

Whatever Agilent/Keysight write, Midi's records demonstrate the ADC aging isn't a random process but something else and very regular. And it does reach 10 ppm/year at a daily drift rate of 0.03 ppm in his case. Hard to believe nobody tried to find out what that process is (and how to stop it).

Regards, Dieter

Why would anyone make the effort when the 3458A is not a primary standard anyway (and not intended to be).
If i want more accuracy for DC i hook up a Fluke 732 and do a differential measurement using the transfer accuracy of the 3458A. Thats a sub-ppm measurement right there, no need to go through the trouble trying to get the 3458A in that region by itself.


edit: Or don't use a 3458A at all, only 732, dividers and null detector.

To protect his investment. If that non-stochastic drift process is the one that finally destroys the ADC, as was argued before. It's nice that you can get a repair, but it's bad if the instrument has a built-in failure mechanism like those Tektronix 2465 scopes i mentioned before.

Regards, Dieter

PS: For example before i buy a 3458A i would ask: Is there a numerical limit in the ACAL procedure? Maybe ADC failure just means that ACAL arrived at a 100 ppm limit or so. Then you can predict how long it will take.

[niner_007]

Which Keysight Lab did you send your 3458A to?

it was Loveland, CO

[Dr. Frank]

...
Whatever Agilent/Keysight write, Midi's records demonstrate the ADC aging isn't a random process but something else and very regular. And it does reach 10 ppm/year at a daily drift rate of 0.03 ppm in his case. Hard to believe nobody tried to find out what that process is (and how to stop it).
....


To protect his investment. If that non-stochastic drift process is the one that finally destroys the ADC, as was argued before. It's nice that you can get a repair, but it's bad if the instrument has a built-in failure mechanism like those Tektronix 2465 scopes i mentioned before.

Regards, Dieter

PS: For example before i buy a 3458A i would ask: Is there a numerical limit in the ACAL procedure? Maybe ADC failure just means that ACAL arrived at a 100 ppm limit or so. Then you can predict how long it will take.

Dieter,
your arguments are very weird.

That small drift does not 'destroy' the U180, or the A/D  at all.
You would probably also not argue, that the drift of gain resistors in other DMMs like in the FLUKE 8508A, 8588A, etc. would finally destroy these instruments.
If you find an excessive drift which would approach or exceed the 24h stability limit, then KS will replace the board, maybe on guarantee under given circumstances, but that's the normal process with failures on any instrument, I think

Also this theoretical drift of 0.03ppm/day will never sum up to 10ppm/year.
Please first read carefully the specification and manual of the 3458A: All specified drift parameters require, that ACAL is used every 24h, or whenever the temperature changes by more than 1°C. 

You might need to understand that the 3458A uses cheaper components, therefore requires frequent ACAL, whereas the DVMs from FLUKE make heavy use of expensive and very stable MBF or hermetic NiCr resistors, to give similar stability figures, but w/o an ACAL process.

The FLUKE 8508A, like the Datron 1281 were much more expensive than the 3458A, and could therefore not compete as industrial use DMMs.

I don't think that these companies buying the 3458A ever feared that their investment was in danger, as a 3458A with a properly working U180 simply fulfills its specification.

If you were still afraid about any small drifts, which are quite natural, then I give you the advise to never buy a 3458A, but also not any other long scale DMM.

Frank 

[dietert1]

@ Frank: 0.03 ppm/day times 365 days/year gives 11 ppm/year. That is an observation of Midi, please look at the diagram posted above. Let's stick to facts.

Regards, Dieter

PS: I guess Midi posted his findings in this thread because he was afraid of loosing his ADC. It's not about me.

[MiDi]

PS: I guess Midi posted his findings in this thread because he was afraid of loosing his ADC. It's not about me.

Essentially no, but I would not consider 0.03ppm/day drift for ADC fully healthy.
Not shure why you got that impression from my posts 
Initially when I got my unit, there was indication of faulty ADC, but after some month of logging this vanished.

[dietert1]

A tracking drift of 11 ppm/year for a 30 year old thin film array appears strange, should be less than 1 ppm. Good ones (still "cheap") are less than 1 ppm when new.
So there could be a deterioration of the ADC silicon and if that is true, it may worry any potential buyer of a 3458A, considering the cost of such repairs. If the deterioration is slow, it will normally be hidden under the ACAL procedure for a very long time, so the problem will only be visible after warranty expired. How much was that "refresh" mentioned above?

Regards, Dieter

[Dr. Frank]

@ Frank: 0.03 ppm/day times 365 days/year gives 11 ppm/year. That is an observation of Midi, please look at the diagram posted above. Let's stick to facts.

Regards, Dieter

PS: I guess Midi posted his findings in this thread because he was afraid of loosing his ADC. It's not about me.

Sorry Dieter,
your arguments are getting more and more weird.
Obviously you really do not understand the specification of, and the technology used inside the 3458A.


The 3458A has a specified drift of 8ppm/year or 4ppm/year for the hi stability reference. This is usually met, because the ACAL procedure every 24h is ultimately required.
The timely drift of the internal A/D resistor network plays absolutely no role. It's also not specified anywhere, as far it's not exceeding the 24h drift specification (defined by hp).
Therefore the ordinaryl U180, indicated by parameter CAL? 72,  might drift 10ppm/year, or much more, but that is absolutely irrelevant, for the 'sanity' of the instrument, for the short to midterm drift, and most important, for the given specification of the instrument.
Latter will stay always below these 10ppm/year, because ACAL must be used.  That's what I wanted to point out initially.
If you would simply judge the state of this instrument from its specification limit, only that would be really 'sticking to the facts'.

Sorry again, but you are arguing about a non specified parameter, which also has absolutely nothing to do with the real, practical use of this instrument.

Additionally, a theoretical 0.03ppm/24h drift is not relevant at all on a daily or even weekly time scale.. please  take into account realistically, how stable analogue DCV standards (eg 732B) are in reality... such a small drift will completely vanish in the noise floor.

Therefore, I would call 0.03ppm/24h for the U180 a really stable and healthy instrument.

Frank   

[MiDi]

Good ones (still "cheap") are less than 1 ppm when new.

Details please, sounds to good to be true 

[dietert1]

No, of course it matters for a buying decision if there are reports about drifting or failing ADCs in 3458A voltmeters, especially since the ADC seems to be the better part of that instrument. You don't have to repeat Agilent/Keysight advertising.

The question for the limits of the ACAL procedure is completely valid and i haven't seen a useful answer. As the existence of this thread shows, there may to be a gray zone, where the ACAL procedure is hiding real problems until it is to late.

As far as i remember 1 ppm is the tracking limit of LT5400 resistor arrays, which have been used by others to make high resolution ADCs. Those are specified with tracking of 0.2 ppm typical. My own experiments with Nomca resistor arrays gave an upper limit on tracking stability of about 0.03 ppm over twelve weeks. That would be 0.12 ppm after a year, if it were a continuous linear drift like your 3458A ADC.

Regards, Dieter

[Dr. Frank]

No, of course it matters for a buying decision if there are reports about drifting or failing ADCs in 3458A voltmeters, especially since the ADC seems to be the better part of that instrument. You don't have to repeat Agilent/Keysight advertising.

The question for the limits of the ACAL procedure is completely valid and i haven't seen a useful answer. As the existence of this thread shows, there may to be a gray zone, where the ACAL procedure is hiding real problems until it is to late.

As far as i remember 1 ppm is the tracking limit of LT5400 resistor arrays, which have been used by others to make high resolution ADCs. Those are specified with tracking of 0.2 ppm typical. My own experiments with Nomca resistor arrays gave an upper limit on tracking stability of about 0.03 ppm over twelve weeks. That would be 0.12 ppm after a year, if it were a continuous linear drift like your 3458A ADC.

Regards, Dieter

Dieter,
Please do not accuse anyone, and especially not me, to repeat advertisement of the companies in discussion.
That's at first not correct, as I simply cited the specifications, and 2nd it's a sign that you've run out of  reasonable arguments.
You're arguing about an assumed failure of the 3458A, which in reality has no real practical relevance.
I'm missing your practical use case, for which this might be relevant, i.e. beyond the specification.
Please explain your use case, otherwise I'd recommend to stop this useless discussion.
Frank

[Kleinstein]

A 0.03 ppm/day drift looks relatively high. However this is not just a single resistor ratio, but there are 2 ratios involved (the  one from the reference scaling about 70% effective). The observed curve from Midi shows some slow down in the drift. This may not be a typical instrument. The ACAL procedure can compensate for such a drift in the resistors, though there are limits.
The feared ADC failure is a different thing, and drift of the gain is only one symptom. With a more frequent ACAL one could probably still use the 3458 with a drift of some 1 ppm/day - however there are other problems waiting to happen, probably trouble with the lesser slopes and thus failure in ADC convergence in the time limits.

The failing ADCs are a weak point for the 3458. There are quite a few reports of such failures. The used instruments from auction sites may include quite some meters that are sold because they showed first signs of the failure or other faults. So there failure rate is naturally higher. There are also the comparators that may fail on the ADC board.

Because of the ADC failures it is important that replacement parts are still available. The old, multi-board-design at least allows partial replacements.  A failing ADC in an 34470 or DMM7510 would be hardly repairable, except for the very skilled volt-nuts. Modern electronics also tends to be more reliable. Maybe we can hope that the black edition also got better in this respect.

[dietert1]

No, i have not run out of arguments. Just wrote down some information to show that there is no good reason why a 3458A ADC should drift at 11 ppm/year. Can't others contribute any information to solve this riddle?

Regards, Dieter