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

[saturnin]

It is suspiciously cheap (for Fluke). Does it include a calibration?

[e61_phil]

It is suspiciously cheap (for Fluke). Does it include a calibration?

Was does "for Fluke" mean? My experience is, that the Fluke calibration is not only cheaper (<1k for 8588A and slightly above 1k for 8508A, but ~2k4 for 5730A) than Keysight, but also much lower in uncertainty. All the Fluke calibrarions had <0.5ppm uncertainty for 10V (last 8508A cal 0.43ppm for example). All the Keysight cal sheets I saw were more in the 3ppm range at 10V. Same with the other ranges.

But you're right. The 1282€ doesn't include the calibration. Together with the calibration it will end very similar to the Keithley 2002 price.

What does that mean for the 3000€ Keysight wants to have? Does it include calibration?

[saturnin]

It is suspiciously cheap (for Fluke). Does it include a calibration?

Was does "for Fluke" mean?

Sorry, I am non-native speaker and used bad expression. I meant that 1282€ is surprisingly low price for 8508A service. I would expect more.  As you wrote, you will end up with a similar price when you add calibration costs. I guess 3000€ Keysight service is the same case, otherwise it is really expensive.

Thanks for overview of calibration costs for other test gear.

[niner_007]

It is suspiciously cheap (for Fluke). Does it include a calibration?

Was does "for Fluke" mean? My experience is, that the Fluke calibration is not only cheaper (<1k for 8588A and slightly above 1k for 8508A, but ~2k4 for 5730A) than Keysight

I don't know about Europe, but 8588A calibration in US is $2,198.00. That was a quote directly from Fluke. 3458A is cheaper to calibrate, with more options available.

[e61_phil]

It is suspiciously cheap (for Fluke). Does it include a calibration?

Was does "for Fluke" mean? My experience is, that the Fluke calibration is not only cheaper (<1k for 8588A and slightly above 1k for 8508A, but ~2k4 for 5730A) than Keysight

I don't know about Europe, but 8588A calibration in US is $2,198.00. That was a quote directly from Fluke. 3458A is cheaper to calibrate, with more options available.

In germany it is 987€ via CalPlus in the Fluke Cologne Lab for an ISO 17025 DAkkS cal. For most cals there is also a cheaper option without DAkkS certificate.

What are the special options Keysight offers?
And I have no doubt that there is a difference in the US. As far as I know Keysight has only a JJA in the Loveland lab. They will offer much better uncertainty than 3ppm I guess.

[ScoobyDoo]

Hello Folks -
                          at a given moment in time - Rigol will throw out its own affordable 8 1/2 - 9 digit DMM ...

                           It is just a matter of time ...

Herzlichen Grüßen/Meilleures salutations/Best regards/Un saludo
ScoobyDoo

[niner_007]

Hello Folks -
                          at a given moment in time - Rigol will throw out its own affordable 8 1/2 - 9 digit DMM ...

                           It is just a matter of time ...

Herzlichen Grüßen/Meilleures salutations/Best regards/Un saludo
ScoobyDoo

It's not just a matter of resolution, it's the accuracy we want I very much doubt Rigol will put out a metrology instrument

[Kleinstein]

9 digits is not rellay realistic (unless 210,000,000 counts would be be called more than 8.5 digits) - more like a slightly faster 8 digit meter.  The difficult part is the linearity, not noise and the reference part is also limiting on how much resultion can really be used. Even with a perfect reference in the meter it is still the usual DUTs that have there own noise.

Rigol likely would not compete for metrological use outside China, but there may be new lower cost instruments in the range of faster 7 digit meters for more industrial use.
With the high prices and limited export from the US, there may still be a place for new competition for high end test-gear from China or similar. Rigol already started at higher grade  scopes.

Service with test gear is a difficult point: the modern instruments got relative robust and do not fail that often anymore, but if they fail, a repair can be really hard, because much is on a single board and highly integrated.  A component level repair is thus difficult and also often expensive unless one knows the faults (like common faults). With the high wages in the US and similar countries such a repair fast gets costly. A board swap looks odd which much on one baord - it may still be OK for the manufacturer, as the actual replacemanet board may not that expensive if one does not include the one time developement costs.
Repair also usually needs recalibration and maybe even a more detailed testing. So it makes some sense to send back to the manufacturer.

With Rigol or Sigilent meter repair is likely also not fun - they have a similar problem and at the lower cost level for the new meter just buying new is even more attractive.  Even just a regular calibration can be a hassel.

The modular design of the 3458 still allows more repairs than with many other instruments. The display and battery backuped RAMs are kind of natural wear parts. Especially for known failure types like dimm display or the dreaded ADC failure a simple board swap without shipping the hole instrument makes absolute sense.

[ScoobyDoo]

Hello Kleinstein, folks - good morning,
                                                     I agree with your post ... - just wanted to trigger the 3458A owners community that KS should keep A3 + A5 + A7 + A9 boards available to end customers as it has been for many years ... and it is good that Fluke is challenging Keysight to further improve the current 3458A - although I doubt that will ever happen ...

Since 2017 - the cost of 03458-66513 A/D board has inflated +3 times (300%). If a skilled technician was capable of servicing such unit in past over 3 decades - why would this no longer be possible in future ? There are many universities that own such 3458A but have problems to afford such standard service repair - the modular 3458A was designed for board level repair ...

In my case I have access to calibration calibres - so I really see no need to return such instrument to KS for a simple board swap and a calibration
session to connect unit to a full automated rack and being charged 3000€ ... - we prefer manual calibration on every range

It just gives me impression that Keysight's management is only concerned to improve shareholders value and not the quality and service level of their product portfolio ... - I hope I am wrong ...

Herzlichen Grüßen/Meilleures salutations/Best regards/Un saludo
ScoobyDoo

[e61_phil]

Hi,

for me as a private 3458A owner it is also a pity that KS will stop selling spare boards, but to be honest, I don't thing it will hit so many companies. In most cases it is way more expensive to let an engineer search for the failure than sending everything to KS, let them fix the box and get a working, tested unit with some warranty back.

And also for the calibration. I doubt that there are many people who are able to calibrate a 3458A by their own (connecting a known 10V and 10k resistor isn't a calibration). On the other side a calibrated 3458A isn't neccessary in most cases. Due to it's superb linearity and low noise it is a very good transfer standard. Absolute accuracy isn't that important for many metrology applications.

[martinr33]

Much of a 3458a is still repairable. The drifty ADC is not. Some of the relays maye be awkward,  and replacing them with sub spec parts is a bad practice.

I don't know if the CPU PALs are protected or not, but it isn't too hard to dump them out.

Displays are available.

Even then, a drifty ADC can be hard to spot quicly without a very stable reference. The recommended process to test drift over a period (assuming you don't have a super stable reference) is someting for which few folks have the patience. Even here, I am not sure how many folks have sprung for a new board.

And one unit I have - early serial number - drifted like crazy for a few weeks but now seems to have settled down to stable. As in, doesn't seem to be mo


One last thing. The 8051 on the ADC board has to be capable of spotting the ADC drift. I wonder if there is some technique that could correct teh meter on the fly. The designers probably did not know that it would be needed. 

[Kleinstein]

Checking for ADC drift does not need a stable external reference. The way to test is checking the cal constant from the ACAL call - so no critical voltage referece involved. Even if the internal A9 ref would drift a lot, it would hardly effect the test. However this test needs time, especially if the temperature is not very stable. A doubt that some of the 8051 are not able to detect the drift - this should be a function available in all versions. For critical use of the meter (e.g. in a CAL lab) it should be good practice to regularly check for the ADC dirft, e.g. like recording the ADC gain value like once a month. A meter not regularly used may show some changes in the ADC gain from humidity or similar effects, not directly related.

I doubt there could be a simple fix (except exchanging the board or U180) to the ADC drift - a failing U180 can cause more problems than just gain drift. The gain drift is only the point that is easiest to measure. If the error gets larger chances are the ADC would complain with a conversion error. It looks like the problem is inside U180 and an external fix is essentially impossible. One can circumvent the problem from moderate gain drift with a more frequent ACAL call, but this would not fix other problems that can come with a failing U180. A drifting ADC may also show higher DNL errors and chances are it will evenuall stop with an error message. Chances are high that when the convergence errors start to pop up that there are extra DNL errors.

There may be a few cases were other parts fail, that could also lead to ADC drift, but this would be rare. There are cases (not drift) of ADC failure that are repairable (though not easy) - especially failing comparators.

[maat]

Even then, a drifty ADC can be hard to spot quicly without a very stable reference. The recommended process to test drift over a period (assuming you don't have a super stable reference) is someting for which few folks have the patience. Even here, I am not sure how many folks have sprung for a new board.

Well it doesn't really take any real patience. Just a Raspberry Pi and a GPIB adapter. Then let the 3458A run for a few days, executing ACAL DCV in a loop and log the CAL? 71, CAL? 71, and CAL? 175 value. The last command queries the temperature at which ACAL DCV was done. It has a higher resolution than the TEMP? command.

Then finally run a linear regression against both time and temperature and you will see where it goes. The SN18 test is a poor man's version of this and if you see a clear linear drift, the numbers given in SN18 are clearly bollocks in this case because their fundamental origin are different (their reasoning and derivation was given by Dr. Frank somewhere in this forum). A faulty U180 also has a higher tempco.

Attached is a sample of a faulty U180. If you want the Python script let me know. I guess this topic is drifting away from the original issue

Edit: Added the Python script

[e61_phil]

Even then, a drifty ADC can be hard to spot quicly without a very stable reference. The recommended process to test drift over a period (assuming you don't have a super stable reference) is someting for which few folks have the patience. Even here, I am not sure how many folks have sprung for a new board.

Well it doesn't really take any real patience. Just a Raspberry Pi and a GPIB adapter. Then let the 3458A run for a few days, executing ACAL DCV in a loop and log the CAL? 71, CAL? 71, and CAL? 175 value. The last command queries the temperature at which ACAL DCV was done. It has a higher resolution than the TEMP? command.

Then finally run a linear regression against both time and temperature and you will see where it goes. The SN18 test is a poor man's version of this and if you see a clear linear drift, the numbers given in SN18 are clearly bollocks in this case because their fundamental origin are different (their reasoning and derivation was given by Dr. Frank somewhere in this forum). A faulty U180 also has a higher tempco.

Attached is a sample of a faulty U180. If you want the Python script let me know. I guess this topic is drifting away from the original issue

Very nice graphs and I really appreciate that you don't give the slope of the fit with an absolute silly amount of digits, but giving some valuable number together with std. dev. 

[maat]

I was asked to elaborate a little on my plot.  I have attached the Python script to the first post and here is what I do:

I do assume there is a linear dependency on the temperature and time. Writing the whole thing in R would probably make more sene, but Python has a nice API to use R-style formulas:

Code: [Select]

from statsmodels.formula.api import ols
model = ols("value ~ numeric_date + temp_deviation", data).fit()
print(model.summary())

Here value is the value of the CAL?72 variable, numeric_date is the number of days since the start of the measurement, temp_deviation  is the deviation from the mean temperature. Now, the model is a multiple linear regression, because the value should depend on both temperature and time. This is important, one cannot do a simple linear regression and either extract the tempco or drift, they have to be extracted at once. The formula simply states, that the column value (left) is related to both the columns numeric_date and temp_deviation (right), and then we want to run ordinary least squares (ols) on it. The next line then prints the summary of the fit.

The script I have attached, can typically easily be extended to work with any data source. Those data sources can be found in the file_parser.py file. Feel free to add your own.
Toggle the boolean plot_tempco to show either the drift plot or the tempco plot.

Edit: A bit of trivia: The A3 board died 3 months later @24/7 usage with a convergence error. So this plot can be considered end of life.

[martinr33]

Some pictures of early U180 devices.
Both have handwritten numbers
The earlier unit has the part information on a sticker on the underside, which explains why some of these are blank.
I used Chipquik to remove one, and am still cleaning up the mess.
The second one, I removed the solde with a solder sucker. This approach did not free the pins. I took it the rest of the way with a hot air pencil, and the package fell out.

I am going to see if baking gets me anywhere.

[Extrasolar]

Hello,

Here are a couple of images from my dead U180 (Resistor Network Ceramic):

[Extrasolar]

Also some images from the Current Steering Section (JFET Switches):

[YetAnotherTechie]

The top edges of the pads look corroded? the line between the pad and the resistor tracks.

[chekhov]

Extrasolar, how you got rid of black resin/coating ?

[Extrasolar]

Extrasolar, how you got rid of black resin/coating ?

Hi Chekhov,

I used Fuming Nitric Acid, that did the job allright (only for the resistor networ) !! The Current Steering Section, needs just a tiny drop of acetone, it disolves rapidly the coating.
I am currently working on a RE attempt on the U180 and the video will hopefully be uploaded sometime in November this year on my channel on Youtube.

Kind Regards

[Extrasolar]

The top edges of the pads look corroded? the line between the pad and the resistor tracks.

Hi,

I think thats just residue from the coating !!

[MiDi]

I am currently working on a RE attempt on the U180 and the video will hopefully be uploaded sometime in November this year on my channel on Youtube.

Link pls 

[Extrasolar]

I am currently working on a RE attempt on the U180 and the video will hopefully be uploaded sometime in November this year on my channel on Youtube.

Link pls 

https://www.youtube.com/user/pitushi/community

[TiN]

Looking forward on your exposure for U180. If you need a hosting for hi-res RAW files or whatever, just ask.