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

[e61_phil]

Also - whan a 3458a is sitting in a semiconductor test rack, I suspect it is using its other skill - super fast lower resolution measurements. The ADC errors won't show up there.

That is also what I think. Most people wouldn't even notice an ADC problem. They will run ACAL before use and everything will stay very well within the 8ppm specification (and a 3458A can do much more than 10V measurements and for the other ranges the ADC drift is even less important). Therefore, I doubt that most 3458As are on eBay because of a drifting ADC. And I also doubt that there was a bad batch in 2006. I saw that problem on much older 3458As.
For me it looks like that problem isn't that unusual and is spread over all 3458A generations. Imho that would also fit to the findings of TiN and ap.

But that was in the days when you could get a replacement A3 board.

If the new A3 board is really 100% compatible, and it sounds like it is, I would guess you can buy new boards for your old instrument soon.

Best regards
Philipp

[Kleinstein]

Also - whan a 3458a is sitting in a semiconductor test rack, I suspect it is using its other skill - super fast lower resolution measurements. The ADC errors won't show up there.

That is also what I think. Most people wouldn't even notice an ADC problem. They will run ACAL before use and everything will stay very well within the 8ppm specification (and a 3458A can do much more than 10V measurements and for the other ranges the ADC drift is even less important).
......
If the new A3 board is really 100% compatible, and it sounds like it is, I would guess you can buy new boards for your old instrument soon.

Best regards
Philipp

The drift is only the first visible symptom of a failing U180. Changes are a drifty ADC will after not so long time (e.g. a year)  show error messages about ADC convergence. The faster conversions are actually more sensitive to the resistor ratios - so DNL errors may be become visible too. I would not be surprised if this would be earlier with the fast modes than with the higher PLC modes.

The reports are that keysight is currently not selling individual A3 boards, but just replace it as part of a repair. This could be to check for possible compatibility issues with older / mixed units. For serious use of the 3458 one would need a new CAL after the ADC board replacement anyway. So sending in the meter for repair is not that much extra hassle for most customers.

[martinr33]

My observation - based on a limited sample - is that there are three classes of drift:

- very stable, close to 0 ppm
- low drift, in spec but clearly observable
- high drift, out of spec

I am thinking, based on your comments,  that there is a fourth class, where drift accelerates out of control. I've not seen that, yet, and my drifting unit was 25 years old    and well outside of the block of units described in the Agilent note.

I wonder if the convergence issue end state is confined to that block of serial numbers.

Also, even a unit with minimal drift can deliver a convergence error if it is never autocal'd.

[notfaded1]

I'd be curious to know what a new A3 with CAL from Keysight would cost?  I'm guessing it's not a normal hobbiest fee?  At least when they sold A3 boards you could buy one put it in and then work a hobbiest CAL with some help from other voltnuts or against your own 10V references and 10K Ohm reference.

Bill

[maat]

I'd be curious to know what a new A3 with CAL from Keysight would cost?  I'm guessing it's not a normal hobbiest fee?  At least when they sold A3 boards you could buy one put it in and then work a hobbiest CAL with some help from other voltnuts or against your own 10V references and 10K Ohm reference.

Bill

Last time (2019) I sent in my 3458A for cal, it is was around 700 €, but I took a deal including an additional year of extended warranty for a bit over 800 €. Replacing parts incurs a fee of about 300 € on top of the part cost (1.5k - 2k for an A3 board). In my opinion sending in the device and taking the warranty extension, which is about 112 €/a is the real deal you can make here, because it will give you the peace of mind, that your little baby will happily live ever after.

Given the fact, that KS collects and returns your device, all in all their service is not that expensive.

[martinr33]

Yes - I think this is one of the things that keeps complete but broken units expensive. Keysight has (had?) that $2700 deal that includes cal and full refurb. So a $3k ebay junker + $2700 gets you a fully refreshed and warrantiable unit.

It is quite possible that they will not make the A3 board available, as other parts have also disappeared. The world has moved on from self-maintenance of instruments.

[dietert1]

Even at that expense they might give you a unit with a drifty ADC and call it "in cal". As far as i understand the official ADC drift limit would be +/- 0.43 ppm/day. And do they borrow you a spare during maintenance? So probably you will end up buying several units.
When buying a pre-owned instrument i would try to get an agreement on a two week test period, like somebody recommended above. The ACAL 72 log of franklin i linked above was in fact only two weeks and it seems to be enough to take a decision, unless you have a "borderline" case like Midi's. And of course i would clean the fan filter, like Midi recommended and also put IC coolers onto those hot comparators. If cleaning the fan filter makes a difference, IC coolers will probably make one, too.

Regards, Dieter

[e61_phil]

The drift is only the first visible symptom of a failing U180. Changes are a drifty ADC will after not so long time (e.g. a year)  show error messages about ADC convergence. The faster conversions are actually more sensitive to the resistor ratios - so DNL errors may be become visible too. I would not be surprised if this would be earlier with the fast modes than with the higher PLC modes.

The reports are that keysight is currently not selling individual A3 boards, but just replace it as part of a repair. This could be to check for possible compatibility issues with older / mixed units. For serious use of the 3458 one would need a new CAL after the ADC board replacement anyway. So sending in the meter for repair is not that much extra hassle for most customers.

I saw 3458As which have a drifting A3 Board and run more than a year with that problem without showing convergence errors or something like that. Even if you send them to calibration no one will notice that. I also doubt, that many people will notice an increased DNL.

If the ADC really develop such problems, it makes the 3458A an unreliable transfer machine. I know many people who are using the 3458A as a ratio machine only and never calibrate it in a traditional way (which wouldn't help). They use it in a single range and measure FS and 0 before a measurement. And I'm sure they aren't looking at CAL72 values.

I hope an ADC which is getting worse in linearity will show that drift, otherwise it is really hard to make sure that your measurements have the uncertainty you expect.

[Dr. Frank]

The drift is only the first visible symptom of a failing U180. Changes are a drifty ADC will after not so long time (e.g. a year)  show error messages about ADC convergence. The faster conversions are actually more sensitive to the resistor ratios - so DNL errors may be become visible too. I would not be surprised if this would be earlier with the fast modes than with the higher PLC modes.

The reports are that keysight is currently not selling individual A3 boards, but just replace it as part of a repair. This could be to check for possible compatibility issues with older / mixed units. For serious use of the 3458 one would need a new CAL after the ADC board replacement anyway. So sending in the meter for repair is not that much extra hassle for most customers.

Kleinstein,
these are very severe assumptions. If these were really true, that would mean that the technology used in the 3458A would not work correctly after some time, affecting then 100% of all instruments.

Please provide proofs for your statements, best from your own experience / instruments, otherwise please give a precise citation where these claimed effects have been observed by others, i.e. here in the forum, or on other platforms.

I do not remember at the moment, that anybody has seen or proven such additional malfunctions in conjunction with a failing U180

Thanks.

[Kleinstein]

The good DNL needs a  good resistor ration for the fast and slow slope. The faster case supposedly need the critical resistor ratio to be good for some 0.1 %. This is normally not critical for a good resistor array. However with a broken chip and dirfty unit (e.g. TiN had some 4 ppm per day), this 0.1% accuracy may no longer hold.  A just borderline unit with 0.4 ppm / day drift would not reach a significant error. The DNL relies on a different ratio - so there is no direct relation saying up to 500 or 1000 ppm total gain drift would be OK.
Depending on the direction of the drift the convergence error may happen (as a waring) or not.

One would hardly notice DNL errors in the slow modes, unless it would happen to be just at zero and cause higher noise.  The slow mode is also less sensitive.  At the high speed a step from, 4.5 digit to 3.5 digit performance may be more obvious and it needs less resistor drift.

Ideally the self test and even the normal power one self-test could check for the resistor ratios at the ADC. This would be directly the fast / slow slope ratio and not the gain drift. However POS is not run with a unit that runs 24/7.  As the software is from the good old times chances are they may do the check, but I don't know.  Today is should be a requirement - but SW quality is more and more diluted with increasing code length. Testing on power up would be still tricky as the meter is not warmed up naturally. I don't know how much self test is included with ACAL - ideally it should do some quick tests too.

Even if just used as a ratio instrument one should still run some checks - the self test on a regular basis (e.g. every 6 months) would be the minimum. Ideally this should include checking the ADC gain (CAL72 value) for the SN18 drift, as a drifty ADC is a known possible mode of failure.

[dietert1]

When researching the web about "3458A U180" i could not find anything from Agilent mentioning quality problems or resistor array drift, except the basic design considerations that Kleinstein repeated.
The only one who really worked on the problem and dared to unsolder U180 hybrids from ADC boards was Tin. He tried to understand the problem better, but did not arrive at a definitive conclusion about what is the typical problem inside a bad U180. I can't understand how a thin film resistor array would cause a tracking drift of several ppm/day.

Concerning firmware quality: Tin described somewhere how to recover a 3458A from a condition where no more calibration can be completed after a previous calibration put the instrument to far off the track. So it's a complicated machine and there might even be a software bug causing calibration to drift or fail on certain drift patterns (for no real reason). But that's speculation again.

Regards, Dieter

[chekhov]

Out of curiosity I've looked into original HP journal dedicated to 3458A design, and resistor network might be not the only thing that this ASIC contains:

Switches
A last major concern for the ADC design was the switches
required to control the inputs and the slopes. Because the
switches are in series with the resistors, they can add to
the temperature coefficient of the ADC. A custom chip
design was chosen so that each switch could be scaled to
the size of the resistor to which it is connected. This allows
the ADC to be sensitive to the ratio-tracking temperature
coefficient of the switches and not to the absolute temper
ature coefficient. Another advantage of the custom design
is that it allows the control signals to be latched just before
the drives to the switches. This resynchronizes the signal
with the clock and reduces the timing jitter in the switch
transitions. The result is a reduction in the noise of the
ADC.

Thus, If I'm reading it right, not resistor may fall but switches.

[Kleinstein]

The main part of the resistance is due to the resistor and only a small fraction (e.g. 0.1% range or less) is due to the switches.  AFAIK HP did not even confirm that the drift is due to U180.
In theory drift of the fets could also lead to gain drift

TiN's measurements did shown that U180 was at fault and they also showed that the reference voltage scaling to +-12 V did drift quite a bit. This part does not include any switches, just 4 resistors from the array.

It is unusual to find such massive drift in a resistor array, but drift already with the scaling part is essentially limited to the resistor part. It also looks that the effect is not just with a few resistors, but more like effecting all resistors to a varying degree.  My best guess would be something like breach of hermetic seal or maybe decomposition of the glue that would release corrosive gas or maybe just the glue coming loose and release of mechanical stress.

[Grandchuck]

My 3458A is in mint condition.  It had been advertised as never used.  Upon receiving it, all original parts and packaging were present.  Not one mark, ding, scratch, or dust in the filter.  But, it seems to fall short of the "golden units" described by TiN and others on the forum.

My voltnuttery has led to the construction/purchase of four LTZ1000 references.  It would seem they are noticeably better than the DMM.  A spreadsheet and graph are included below.

The DMM uncertainty (36.4 uV @ 10V) is specified in the calibration report that was part of the package.  The meter is branded Agilent and the date of manufacture is 2011.

[TiN]

He tried to understand the problem better, but did not arrive at a definitive conclusion about what is the typical problem inside a bad U180.

I did arrive to definitive conclusion, but I thought it's obvious enough already and didn't feel repeating it. If one cares for meter to meet spec, SN18 is official as it gets. If one wants better than specs, well, you are on your own hunt for magical 0.0ppm/yr system.

Feeling like every year people here refresh and have this kind thread about u180, repeating same things over. I remember Dr.Frank explained these same things at least 3 times by now. 

[dietert1]

We have been using HP 3456A for years and years. Those don't have that ACAL procedure to check the ADC, so all ADC deviations enter as errors into the measurement. The ADC has to be very stable and it is (except some small temperature coefficient). Long term stability is something like +/- 1 or 2 ppm per year. Our instruments are old and well aged.

Reading this forum one may get the impression that a 3458A is a must-have for metrology, because it's ADC has the best linearity. But then we read that its ADC contains a part that may cause tremendous drifts. Drifts up to 157 ppm/year are officially considered "normal", as long as the ACAL procedure takes care. So in my eyes the 3458A series is a similar kind of retarded progress as the 3457A series. Just think about the fan that came back after it vanished from the 3456A and the 3457A.

I think people are asking over and over again, because all this is hard to believe and difficult to understand. What i take from this discussion is that a metrology grade 3458A is something special that money won't buy. Maybe that's the reason why people try to be friendly with Agilent/Keysight.

Regards, Dieter

[The Soulman]

My 3458A is in mint condition.  It had been advertised as never used.  Upon receiving it, all original parts and packaging were present.  Not one mark, ding, scratch, or dust in the filter.  But, it seems to fall short of the "golden units" described by TiN and others on the forum.

My voltnuttery has led to the construction/purchase of four LTZ1000 references.  It would seem they are noticeably better than the DMM.  A spreadsheet and graph are included below.

The DMM uncertainty (36.4 uV @ 10V) is specified in the calibration report that was part of the package.  The meter is branded Agilent and the date of manufacture is 2011.

What is the timing interval of those measurements?
Did you use autocal prior to each measurement?

[Dr. Frank]

He tried to understand the problem better, but did not arrive at a definitive conclusion about what is the typical problem inside a bad U180.

I did arrive to definitive conclusion, but I thought it's obvious enough already and didn't feel repeating it. If one cares for meter to meet spec, SN18 is official as it gets. If one wants better than specs, well, you are on your own hunt for magical 0.0ppm/yr system.

Feeling like every year people here refresh and have this kind thread about u180, repeating same things over. I remember Dr.Frank explained these same things at least 3 times by now. 

Illya,
thank you for your explanations, setting the discussion about drift parameters of the 3458A to the correct, practical perspective.
I hope that this will finally calm down some owners of a 3458A, who were made nervous by several theorists here - sorry I can't call them otherwise.

Latter, who even compare a 6 1/2 digit DMM to the 3458A in terms of stability or alike, thus obviously demonstrating that they did not even understand principle and specification, I really can not take serious any more.


Frank

[Dr. Frank]

My 3458A is in mint condition.  It had been advertised as never used.  Upon receiving it, all original parts and packaging were present.  Not one mark, ding, scratch, or dust in the filter.  But, it seems to fall short of the "golden units" described by TiN and others on the forum.

My voltnuttery has led to the construction/purchase of four LTZ1000 references.  It would seem they are noticeably better than the DMM.  A spreadsheet and graph are included below.

The DMM uncertainty (36.4 uV @ 10V) is specified in the calibration report that was part of the package.  The meter is branded Agilent and the date of manufacture is 2011.

Hi Grandchuck

It's not clear, under which conditions you made your measurements, i.e. NPLC setting, temperature stability of room / 3458A, frequency of measurement..
I have just recently published such stability measurement of a 3458A vs. LTZ1000 reference (or similar).
 +/- 0.1ppm over 24h, 200nV short time noise shall be achieved for healthy devices.

Maybe you try to make a similar measurement.
From your data, I can't realize how you can state that your 3458A  does not perform as well as the LTZ1000 reference.

Frank

[The Soulman]

We have been using HP 3456A for years and years. Those don't have that ACAL procedure to check the ADC, so all ADC deviations enter as errors into the measurement. The ADC has to be very stable and it is (except some small temperature coefficient). Long term stability is something like +/- 1 or 2 ppm per year. Our instruments are old and well aged.

Reading this forum one may get the impression that a 3458A is a must-have for metrology, because it's ADC has the best linearity. But then we read that its ADC contains a part that may cause tremendous drifts. Drifts up to 157 ppm/year are officially considered "normal", as long as the ACAL procedure takes care. So in my eyes the 3458A series is a similar kind of retarded progress as the 3457A series. Just think about the fan that came back after it vanished from the 3456A and the 3457A.

I think people are asking over and over again, because all this is hard to believe and difficult to understand. What i take from this discussion is that a metrology grade 3458A is something special that money won't buy. Maybe that's the reason why people try to be friendly with Agilent/Keysight.

Regards, Dieter

In my mind the 3458a ADC had the following design goals:
*high resolution/dynamic range
*excellent linearity
*high speed

And with those three things combined there was no room for:
*long term stability
*low power consumption/low temperature
*low cost
*numerous other things

The long term stability has been "solved" by using autocal frequently, that is what it is there for.

[The Soulman]

From the manual:

When to Use Autocal
For maximum accuracy, we recommend performing ACAL ALL once every
24 hours or when the multimeter's temperature changes by ± 1°C from when it was last externally calibrated or from the last autocal. (We recommend that the calibrator store the multimeter's internal calibration temperature using the CALSTR command; this can be read later using the CALSTR? command.)

[Grandchuck]

More information about my measurements:

1/ NPLC 100
2/ Readings taken every 24 hours
3/ DMM warmed up at least 4 hours before each reading
4/ DMM turned off each day
5/ Autocal performed immediately before each session
6/ Lab is temperature and humidity controlled (plus/minus 1 degree C)

Thanks Soulman and Dr. Frank!

My conclusion is that the DMM internal reference is moving a bit ... perhaps due to the on/off cycling.  The external LTZs are powered constantly.  Autocal shows little to no change.  That would indicate that the A/D board is solid? 

TiN, how is "superstable" LTZ1000 reference project coming along?

[Villain]

More information about my measurements:

1/ NPLC 100
2/ Readings taken every 24 hours
3/ DMM warmed up at least 4 hours before each reading
4/ DMM turned off each day
5/ Autocal performed immediately before each session
6/ Lab is temperature and humidity controlled (plus/minus 1 degree C)

Thanks Soulman and Dr. Frank!

My conclusion is that the DMM internal reference is moving a bit ... perhaps due to the on/off cycling.  The external LTZs are powered constantly.  Autocal shows little to no change.  That would indicate that the A/D board is solid? 

TiN, how is "superstable" LTZ1000 reference project coming along?

I know for some this sounds stupid but when i hear someone saying "turning off 3458A each day" my stomach twists and turns.

[Dr. Frank]

More information about my measurements:

1/ NPLC 100
2/ Readings taken every 24 hours
3/ DMM warmed up at least 4 hours before each reading
4/ DMM turned off each day
5/ Autocal performed immediately before each session
6/ Lab is temperature and humidity controlled (plus/minus 1 degree C)

Thanks Soulman and Dr. Frank!

My conclusion is that the DMM internal reference is moving a bit ... perhaps due to the on/off cycling.  The external LTZs are powered constantly.  Autocal shows little to no change.  That would indicate that the A/D board is solid? 

TiN, how is "superstable" LTZ1000 reference project coming along?

Yes, measuring once a day, but shutting off in between is not the best method.

Anyhow if you 're using the 3458A intermittently, like on a monthly basis, like here: https://www.eevblog.com/forum/metrology/hp-3458a-ad-board-spare-part-availibilty-and-cost/msg3156428/#msg3156428
Then these fluctuations should be 1/10 of yours only, i.e. +/- 0.2ppm from day to day. Especially this common dip +2ppm at day 6 makes me wonder, what's going on.

You also need to take each measurement with sufficient statistics, i.e. with standard deviation, so you can see if something is fishy.
Personally, I take 16 readings @ NPLC 100, with statistics, for each reference and for a single measurement point.
If the StD of 3458A vs. LTZ1000 is from 100nV to 300nV, then the measurement is ok.
You should also check if your used cable is ok, i.e. low thermal gold plated or pure copper. This can be checked by reversing the cables at the DUT and comparing if + and - measurement are identical to about or below 1µV.

I would also recommend, to do a 24h stability monitoring , 3458A vs one of your LTZ1000, and with TEMP? monitoring, whether you get such stable curves like mine.

From that you can very often deduce, what's wrong.

Frank

[Grandchuck]

Thanks Dr. Frank.  The 3458A has been on continuously for a week and things are looking better. 

So, this is an instrument that should not be turned off when seeking best metrological performance ... makes sense, I suppose