3458A and ADC drift

[niner_007]

Hi, I'm trying to understand why some of the 3458A bough from ebay have a bad ADC board which drifts, is this something affecting some units (US28031400 - US28032927 according to SN18) and only a problem in the early units, or could this potentially happen to a modern 3458A manufactured in the last 1 to 2 years? Or is the ADC hybrid in 3458A intrinsically problematic? 

[TiN]

Are so many 3458 threads really needed? Could just put all your questions in one, and call it "3458A discussions" or something.
U180 is weak spot for reliability, even new 3458A had cases when ADC failed/started to drift. I have 100% luck with U180, meaning all are faulty or drift more than 0.01ppm/day.

[niner_007]

Thanks! TiN, do you have a modern 3458 that is drifting, or saw it elsewhere?

Regarding thread count, I personally find it easier to find/distill information when it is compartmentalized, as opposed to one massive thread. But I can follow the forum rules if that is the recommendation.

[Kleinstein]

The drifting ADC (usually U180 failure) is a common problem.  Chances are with many of the units seen on Ebay a no longer good performance already showed up before they where taken out of service. So the sold units are not a random sample, but to a large part likely know problematic units.  They may not know the SN18 service note and just see that they are relatively drifty and thus decide to sell.

The question whether the drifty ADC only effects old units is a good one - but I don't know an answer.  At HPAP they likely know more details, but I don't know if they have changed the ASIC (or something else) to avoid such failures in newer units.

[niner_007]

The question whether the drifty ADC only effects old units is a good one - but I don't know an answer.  At HPAP they likely know more details, but I don't know if they have changed the ASIC (or something else) to avoid such failures in newer units.

Based on SN18s serial number range, one could be led to belive that there was a manufacturing defect with some units U180, and they addressed it, but that might not be the case indeed.

[martinr33]

I've seen units with bad ADCs outside of the serial number range. I have two drifty boards. One is dated 1990, and one is dated 1998.

The official affected serial number range is US28031400 / US28032927. However, the 1990 unit is too early for this range.

A good board drifts so little that it is hard to detect. Autocal actually throws in discontinuities in measurement. Some units drift a little, enough that autocal is useful.

Buying a 3458a used, you have to figure that you are risking a $1200 refurb on just the ADC (although there are some refurb units availabel for less on occasion).

[TiN]

Yea, I wouldn't pay much attention to "affected" serial numbers, as none of my boxes and boxes of friends that I know about were in the line of "affected", yet still bad/drifty A3's (10+).
Also "limit" 0.43ppm/day in SN18 is ridiculous, actual good A3 should be limited to no more than 0.01ppm/day, otherwise meter will not meet it's annual specification properly. Using ACAL is not a solution for ADC drift, because drift is only obvious issue, but it's not the only problem that drifty ADC U180 chip have. 

[MiDi]

Using ACAL is not a solution for ADC drift, because drift is only obvious issue, but it's not the only problem that drifty ADC U180 chip have. 

Only "high" TC comes to my mind, are there other problems?
Edit: in addition my mind told me there could be linearity problem.

[HighVoltage]

The CAL company I use here in Germany also repair 3458A a lot and the guy told me that even very new Keysight branded 3458A can show a drift and bad A3.

[IanJ]

Hi all,

For info........here's my Cal?72 records thus far.....includes a recal at Keysight so I reset the calc then (day 157).......I presume it makes sense to calc based on Day 157 and not keep based on day 1...?

3458A manufactured Feb. 2000. (S/N = 2823A24074).
Powered 24/7.

285 days in total....scroll the data to see it all.

Code: [Select]

Day      Date              Time            GPIB      Temp.       CAL? 72 (SN18 Drift)        CAL? 1,1 (40k)    CAL? 2,1 (Vref)     Drift PPM (Ref to Day last 'Start')
1        11/10/2018        9:00:00 PM      22        36.4        0.984806743                 39.9982551        7.07046210          Start
2        12/10/2018        1:50:00 PM                36.7        0.984806756                                                       0.0132
3        13/10/2018        9:00:00 PM                36.8        0.984806811                                                       0.0690
4        14/10/2018        7:00:00 PM                36.4        0.984806762                                                       0.0193
5        15/10/2018        10:30:00 PM               36.3        0.984806911                                                       0.1706
6        16/10/2018        8:00:00 PM                36.5        0.984806882                                                       0.1411
7        17/10/2018        9:00:00 PM                36.5        0.984806948                                                       0.2082
9        19/10/2018        4:00:00 PM                35.9        0.984806838                                                       0.0965
10       20/10/2018        11:00:00 AM               35.8        0.984806776                                                       0.0335
11       21/10/2018        6:00:00 PM                35.7        0.984806654                                                       -0.0904
12       22/10/2018        8:30:00 PM                35.4        0.984806750                                                       0.0071
13       23/10/2018        7:30:00 PM                35.6        0.984806823                                                       0.0812
17       27/10/2018        8:00:00 PM                35.6        0.984806903                                                       0.1625
20       30/10/2018        8:00:00 PM                36.0        0.984807056                                                       0.3178
78       27/12/2018        11:00:00 AM               38.0        0.984807203                                                       0.4671
128      15/02/2019        16:00:00 PM               36.2        0.984806808                                                       0.0660
157      16/03/2019        16:36:00 PM               35.3        0.984806200                                                       Start (Recal at Keysight 08/03/19)
157      16/03/2019        20:34:00 PM               35.3        0.984806212                                                       0.0122
160      19/03/2019        21:40:00 PM               35.2        0.984806139                                                       -0.0619
193      21/04/2019        19:30:00 PM               36.6        0.984806476                                                       0.2803
193      21/04/2019        20:00:00 PM               36.6        0.984806473                                                       0.2772
285      22/07/2019        18:30:00 PM               38.2        0.984806861                                                       0.6712
Today w/shop is quite warm at 26degC hence 38degC 3458A temp today. I notice at the 2 days it was 38degC (Day 78 & 285) the PPM was at it's highest.....but then recovered back down again when temps returned to normal

Ian.

[Dr. Frank]

Hello,
my opinion is, that a 'bad' U180 ADC hybrid is not defined well, neither by this infamous SN18 (too crude criterion, also incomplete), neither by KS, or any cal lab..

My main point of criticism is, that this CAL? 72 parameter definitely depends on the environmental temperature, which is not mentioned at all in SN-18.

CAL? 72 indeed mostly reflects, or compensates this temperature related drift.
That is the T.C. mismatch of the TaN (or else) R-network inside the 3458A.
It is indeed specified to be around  0.5ppm/K, including the LTZ1000A T.C., latter is specified 0.15ppm/K, so leaving about 0.35ppm/K for U180. My 3458A shows about -0.24ppm/K for the U180.

But that is exactly in the same ballpark as the SN18 criterion, so it's necessary to separate all these three effects from each other, i.e. U180 network T.C., LTZ1000A T.C., and possible (excessive) drift of resistors inside U180.

Frankly speaking, I just can't believe, that Illya has 3 different repaired 3458As, which all should have zero T.C. for their LTZ1000A reference, as well as for the U180.. Sorry Illya..

A more practical characteristic, or more typical instrument like mine, really shows about 0.4ppm/K overall T.C., and no discernible drift (apart from temperature effects) concerning U180 resistor network ( SN18).

I attach the recent tracking ..
Please take notice, that the temperature of the U180 is very un-precisely determined by the TEMP? value, as the temperature sensor and U180 are sitting top and bottom inside the instrument, so that explains  these big variations at a fixed temperature.
The single event  in about December 2018 (+2.75ppm) reflects an ACAL DCV performed directly after switching on, that is the point at 25°C.
I also did a new calibration recently, after import from our Stuttgart Metrology Meeting, so that's the +1.8ppm jump (after about 10 years!) of the LTZ1000A value, which also goes into the CAL?72 value, as Ian already mentioned.

Conclusion: I propose to make some more detailed investigations (over temperature) on a specific 3458A, before one claims it to have a 'faulty' A3 board!
Frank

PS: Something additional: the ACAL DCV behavior - especially on sub ppm level and sub specification parameters- depends on, whether you use/power the instrument intermittently (like in my case), or let it always on 24/365, like in Illyas case.
My instrument shows ACAL DCV jumps on the order of 0.2..0.3ppm, which I mostly assign to internal noise.
That's all covered by the specification, so I also would be critical about statements, that this is an unusual / defect behavior..

[TiN]

My main point of criticism is, that this CAL? 72 parameter definitely depends on the environmental temperature, which is not mentioned at all in SN-18.

Yes, but it's "easy" to find correlation once TEMP? data is collected together with ACAL cycles before/after recording CAL? data. I run this procedure on somewhat random period, here's plot of CAL?72 for my "rusty" 3458A. Be my guest to find temperature correlation in it

Frankly speaking, I just can't believe, that Illya has 3 different repaired 3458As, which all should have zero T.C. for their LTZ1000A reference, as well as for the U180.. Sorry Illya..

No need to apologize. But you don't see the complete picture. I've repaired more than 8 3458A's, and my own 4 meters, and over went thru over 10 A3's. So far my only two, not three meters are 0.0x ppm/K TC, rest are more usual 0.2-0.3 ppm/K. I think we have over 20 meters in our little xDevs chat team to collect data from, if you'd like to join our talk.

I believe we are going over this again, just like number of previous diluted 3458A threads . I do checks almost every month on my 3458s, because I cannot believe 0.0x ppm/K TC myself too, perhaps. Here's July data over 3 days, with two excursions from +23 +/-2°C to +29 +/-2°C. Ambient room temperature is black line (monitor by BME280).



DSV-source file with all points. DSV file also have TEMP? readings for all meters, which are not shown on the plot above. All equipment state in lab is unchanged (to remove doubts that heat gradients during test changed due to other gear around).

All seven meters are connected willy-nilly with low thermal cables to AC-powered Fluke 732B reference (calibrated vs PJVS at May 31, 2019).
Legend:
3458-2 - my first "rusty" meter, replaced A3 ADC 3 times.
3458-3 - my second meter, replaced A3 ADC 1 time to 2016 year one. Meter is modified, have lower DCV noise.
3458-10 - 4th meter, came with 1ppm/day drifty ADC. replaced A3 ADC 2 times.
3458-11 - 3th meter, came with 0.1 ppm/day drifty ADC. installed A3 ADC 1 time.
K2002-6 - one of the 2002's, no modifications. My other 2002 have much better TC, but was busy with other experiment at a time.
D1281  - fresh addition to the team, old but mighty Datronski. TC is so-so.
F8508A - nothing to explain here. Good TC , but worse than my "mythical" 3458s.

Worthy note - both 3458-2 and 3458-3 have modified A9 reference. -3 unit have LTZ1000CH (not ACH) and 95kOhm S102 in parallel with 15K. -2 unit have original LTZ1000ACH chip, and 100 kOhm VAR BMF in parallel with 15K. -11 and -10 run stock A9's so far, with cooking LTZ temperatures.

There is also second test, with even warmer extreme, with 2 days of data.



DSV-source file with all points

Same test condition. Test is not highly scientific, proper way would be to keep 732B (or the meters) in the isothermal climate chamber, but I don't have anything big enough to fit 732B, so we make assumption that my ref meets it's factory specification <0.04 ppm/°C for 10V output. I have no much reason for doubt here, because my previous tests with xDevs FX reference in constant +23.00 °C oven show same results over 3458A/B (-3, and -2 units), so data correlate well.

Also my other past data show evidence that 10V range TC is not 100% relate to A3 ADC only, I have enough units and A3 boards to prove see that now, but I'm not ready to disclose details about this yet, because this is part of bigger project study. Sorry about that.

I propose to make some more detailed investigations (over temperature) on a specific 3458A, before one claims it to have a 'faulty' A3 board!

While I do not disagree, that testing ultra-high performance DMM/ADC like 3458A can be challenging task, it is not that hard to sort faulty A3 vs good A3. If meter (as a whole) fails meet it 24 / 30 day specifications with ADC A and it meets them with known good ADC B (with guardband margin, let's say 60% just to be sure) - then there is nothing else to discuss, and ADC A is faulty. Yes, there is variation from reference noise, like Dr.Frank correctly highlights, but it's irrelevant even over few weeks data, leave alone a month.

Hence main conclusion is that one cannot just spend a day or two with 3458A and determine if it's faulty or good (as a whole), making all those ebay listings for "pulled from working environment, NIST calibrated" not worth too much. For lab on a budget (he-he) best way is to have known-verified good 3458A as a "reference" unit and do verification of unknown A3 ADC/A9 refs in it, so things can be tested one at a time, instead of trying to figure out if unknown instrument as a whole is good or not.

[niner_007]

3458-2 - my first "rusty" meter, replaced A3 ADC 3 times.

TiN, I lack context here so if you said it elsewhere, feel free to point me to it, but why did you have to replace A3 three times? Did you not source the A3 from Keysight, or had to replace it three times over time? If you did not source it from Keysight, then I can only say you were unlucky and I don't think it's indicative of anything, whether 10 or 100 replacements.

[TiN]

Well, if you think that Keysight sells new A3 or A3 that are tested for our tighter drift expectations, you are in for a surprise. And remember, I am not in the business of T&M , all above is just personal little lab observations, from data results.

[niner_007]

I would have hoped so Worse, they don't sell new A3s anymore

[martinr33]

One of the things that I have noticed is that ACAL can cause a bigger deviation than the thermal shift on some units.

One day, it would be nice to characterize the thermal drift and comp it out.

TiN - do you think that low thermal drift you have comes from the analog side, the ADC, or the LTZ1000?

[Kleinstein]

The Ref. circuit modification to a lower set temperature is done to get less long time drift. It has a small effect on the TC of the reference, but it is not clear if improving or not.  I would consider the exceptionally low TC  for the 2 units more or less good luck. Chances are the A3 and A9 both have some TC and there is some compensation.

The drift test just needs time to distinguish between a possible temperature and aging effect. So it's more like 12 ppm over a month, not 0.4 ppm over a day.  The bigger trouble may be the "normal" drift one may see in units that where out of service for a linger times - this may need extra time to see if the drift is from a bad U180 or just normal bake out.

If the A3 availability is so bad one may have to consider the Keysight extended warranty as part of the costs maintaining a 3458.

[IanJ]

I would have hoped so Worse, they don't sell new A3s anymore

When I bought an A3 from Keysight UK last year it was stipulated it may be brand new or may be a refurbished board.

Ian.

[niner_007]

I would have hoped so Worse, they don't sell new A3s anymore

When I bought an A3 from Keysight UK last year it was stipulated it may be brand new or may be a refurbished board.

Ian.

Interesting. They don't sell new A3s anymore, they sell refurbished, and they explicitly advertise it as such. Both come with "tested" tag, but not sure what it means.

Yours was drifting too, or had other issues?

The price difference is almost double, $800 for refurbished vs $1,400 for new

[IanJ]

I would have hoped so Worse, they don't sell new A3s anymore

When I bought an A3 from Keysight UK last year it was stipulated it may be brand new or may be a refurbished board.

Ian.

Interesting. They don't sell new A3s anymore, they sell refurbished, and they explicitly advertise it as such. Both come with "tested" tag, but not sure what it means.

Yours was drifting too, or had other issues?

The price difference is almost double, $800 for refurbished vs $1,400 for new

Not my current 3458A, it was a 3458A I bought cheap from overseas as faulty......turned out the A3 wasnt even powering up without fault, and would freeze after a few minutes. A replacement A3 fixed the problem.

Theres a whole thread here somewhere with the repair log.

Ps. What I would love to see is a re-engineered solution for the U180......not by Keysight though, but by a third party. Anyone tried, or thought about trying?

Ian.

[niner_007]

Here: https://www.eevblog.com/forum/metrology/agilent-3458a-repair

[TiN]

Ps. What I would love to see is a re-engineered solution for the U180......not by Keysight though, but by a third party. Anyone tried, or thought about trying?

Help yourself. 

[MadTux]

That silicon on top is most probably, where the problem with reverse engineering is.

"Special parts" on older HP meters, 3455A, 3456A are purely resistor arrays, no big deal building replacements from discrete parts, there. Reversing the silicon is a lot harder. That's why I stick with older, simpler stuff, much easier to maintain, if the added accuracy is not needed.

[Kleinstein]

----
Ps. What I would love to see is a re-engineered solution for the U180......not by Keysight though, but by a third party. Anyone tried, or thought about trying?

Just making a U180 replacement would still need a new linearity qualification and so one. So there would be big extra hurdle to to get a replacement at that level. As there are different A3 versions around and they may behave different there is very little hope to get a direct 1:1 replacement for U180. The quantity for a replacement would also not be that large - not every U180 fails. For a one off unit - there is some hope the silicon is still OK in some units and the lower part only contains a bunch (quite a few) of high end precision resistor. Still parasitic capacitance still make a replacement nearly impossible.

I thought about a complete A3 board replacement - but the fast conversion modes are difficult to recreate with a different type of ADC. The interfacing to the AC board can also be tricky.  Chances are a complete new DMM may be easier than trying to replace only the A3 board and keep the rest.

I think it is absolute time for Keysight to get a new design (not just new color). There are more problems than just occasional U180 failures: The 3458 is large, power hungry uses old SRAMs with integrated battery, no LAN, expensive to build and it used old custom parts (e.g relays, resistor chips, ...). The fast digitizing performance and digital AC is no longer that unique. A new design very likely would no longer use a U180 equivalent chip, but a new ADC or two.

[niner_007]

If they are going to do a patch, they won't invest a lot of resources into it, I agree if they have to, they would just redesign the entire DMM. I don't think we saw a new design since K2002, all Fluke 8.5 DMMs are just iterations over Datron, same ADC? 8.5 DMMs are needed in the industry, but we might never see another 8.5 DMM that is affordable, and based on a brand new design. Since it's so niche, why not sell it at $100,000 or $200,000, I expect that will slowly happen in today's economy.

[Kleinstein]

In industry there is little need for 8 digit resolution. The big point of the 3458 there is more in getting good resolution (5-7 digits) fast. However in this area some modern meters like the Keithley DMM7510 can be nearly as good and an optional scanner is a big plus.
The real need for 8.5 digits is essentially only in metrology, e.g. to check other meters / calibrations and similar. So it's a niche market and it's much about reputation.

At least from the specs the 2 new Fluke 8.5 digit meters seem much lower noise than before. So this may very well be more than just a simple slight upgrade of the old Datron ADC.

[Dr. Frank]

Well, the upcoming '3458A - Black Edition' will probably contain nearly the same schematics inside as the latest version, but most of its components will be transferred to SMD and RoHS technology only, therefore smaller PCBs, plus replacement of several terminated parts. Hopefully, they did not touch the magic inside the original layout..

Everything might be compatible to the old instruments, as HP /agilent already did that during the last 30 years.

Maybe there will be an improved specification, as KS surely has to go through a new validation / qualification process, which  might be the reason, why the updated instrument is not yet available.

I also assume, there will be no firmware change. It'll have the same GPIB interface only... HP did also not want to touch the rock solid firmware of the HP48 calculators, later on.

KS simply does not have these 'eggheads' anymore, who either left the company (towards Keithley), retired, or already passed away (poor Mr. Swerlein).

Some remainders of the original team just designed the 34465A/470A, but the knowledge about that 3458A magic surely passed away also, and I really doubt that this had been documented well enough.

Although there would be some details to be improved, like this mistreated LTZ1000 reference, the thermal handling, the poor Ohm mode, or some components which have too big a drift.. but why change a near perfect design, especially this unsurpassed A/D converter, for no real advantage, no real perspective for noteworthy improvement, due to probable physical limitations?

Frank

[iMo]

.. but why change a near perfect design..

Why ElonM and his SpaceX does not use the Sojuz or Proton rockets then? 

[niner_007]

Proton and Soyuz rockets are far from perfection

[Zucca]

Well, the upcoming '3458A - Black Edition'

we should start a separate thread only about the Beast Edition 3458A. Right now we have some informations spreaded in different places.

[Gandalf_Sr]

.. but why change a near perfect design..

Why ElonM and his SpaceX does not use the Sojuz or Proton rockets then? 

Elon Musk has ALL the key stuff designed and built in-house; from fuel tanks to triple-redundant avionics.  For one thing, buying rockets from Russia would seem like a very dodgy proposition for a company that's contracted to deliver secret stuff into space for the US government and anyway, reliance on an external contractor is always a risk that you will struggle to contain.

I truly admire Musk for what he's achieved but I'm not sure I'd like to work for him.

[iMo]

FYI - my question above was rather a rhetorical one

[niner_007]

FYI - my question above was rather a rhetorical one

yes we know, the SpaceX is Keysight here

[maat]

Well, save the date, 1st of December:
https://www.meilhaus.de/en/3458a.htm

[iMo]

Any teardown seen?

[niner_007]

Well, save the date, 1st of December:
https://www.meilhaus.de/en/3458a.htm

The king is dead, long live the king :-) The beast is back.

[rogerymw]

Hello Everybody,

I just testing an old "HP" Logo 3458A, the CAL? 72 data as attachment.

A is time : Mins
B is Temp? value
C is CAL? 72 value


Is that the drift too high?

[Kleinstein]

I just testing an old "HP" Logo 3458A, the CAL? 72 data as attachment.

A is time : Mins
B is Temp? value
C is CAL? 72 value

Is that the drift too high?

It depends on the unit for the times scale:
For this is time in seconds the measurement is too short. If this is time in hours or days  , the dirft would be OK. For a time in minuts the drift is too high. It could still be just a settling effect after longer not using the meter - in this case it may still stabilze and need a little more time to tell.

[rogerymw]

off about a month before.

how about yours? for how long it will drift less?

Thanks!