Repair and experiments thread: HP 3458A U2

[Kleinstein]

Filtering of the reference noise is possible for the higher frequency range (e.g kHz), where many DMMs are also somewhat sensitive, but this not such an important issue and many DMMs don't even use filtering here.
For the below 1 Hz range filtering is not practical, just because resistors / caps get to large. However it is possible to use separate references for the short time (e.g. up to hours / days) and long time ( > days) time scale. Many DMMs have that naturally (though they usually still use the same ref.): one for the ADC and one to measure directly in the 10V range to adjust the gain. So a low noise DW232 could be combined with a log time stable LTZ1000. The Keithley 2001 goes the way of two references - though I am not sure it is implemented very well.

Comparing ref and AZ OPs noise is difficult, as AZ noise is more or less white, while the reference noise usually has quite some 1/f component. So in the 0.1 to 1 Hz range things are much more in favor of the AZ-OPs.

Clock Jitter is the least problem - so a normal crystal clock should be good enough even for 9 digits. On the critical short time scale a GPSDO is not better than a crystal anyway, as they internally use a crystal as well. The kind of phase noise such an integrating ADC is sensitive is more like phase modulation in the 10s of kHz - not frequency drift in the 10s of ms range. The problem is not so much measuring the time, it is more about parasitic currents and nonlinearity due to higher order effects like output voltage dependent bandwidth, ringing after fast switching or nonlinear resistors and just simple current noise of BJT based OPs. Also nonlinearity and dielectric absorption in the integrating capacitance can become an issue - this is why such a small capacitance is preferred, despite of more frequent reference current switching and thus higher sensitivity to clock jitter.

[Theboel]

according to "High resolution digitizing techniques with an integrating digital multimeter by David a czenkusch" You can read it with "HP 3458A Component-Level Information Packet.pdf"
3458A use 10Mhz as reference and the clock has less than 150PS at 10KHz jitter if somebody can measure this and compare to better clock we can now find out what the result with a better clock.
the other thing in the paper are talking about are capacitor accuracy 0.001pF 
so my list now get longer :
1. use 4 LTZ1000 (or more) in paralel to bring the reference noise down
2. Use temperature control system to keep everything inside has a stable and low humidity especially reference and ADC (yes I know about ACAL but if I keep everything 23 degree with 1/10 degree variation there is nothing wrong ?)
3. Better PSU and filtering.
4. Better clock
5. Better capacitor divider network.
sorry if I made mistake take my opinion as dummies thinking about the "king of the hill" like TiN said

[TiN]

CalMachine

I'm skeptical about "magnitudes" of improvements possible in any key performance aspects. Even for testing alone, hope you get that programmable JJA system in place, as that's pretty much only way to really split beans during testing whatever ideas we try. As was already mentioned, even if we design and get 9.5 digits, they mean nothing (other words - just noise) without linearity and stability to backup. And testing sub-ppm linearity is already hard enough, even with PJVS.

Having that said, replacement of onboard oscillator with more stable clock module is one of the things we can try in few months, once I confirm 3458B is fully stable and functional in stock condition. I'd expect rubidium-based source might be a better option with less phase-noise compared to GPSDO+OXCO combo. I have both anyway.

Btw, so far new ADC is good and stable, -0.01ppm drift over 7 days per SN18 procedure.

I also did a side test for 10V, 100V, 1000V ranges using same 10V reference.
To confirm the numbers, I ran same test with my reference 3458A using 10V out of 3245A (+/-0.5 ppm stability).

3458B 10V range Deviation	3458A test	3458A 24h spec
+0.25ppm, Reference, 10V range	0.0 ppm 9.999978 VDC reference from 3245A	+/-0.275 ppm
-1.6ppm 100V range	-1.5 ppm Datalog 100V range	+/-14 ppm (+/-3 ppm for 10 min )
+6.0ppm 1000V range	+3.4 ppm Datalog 1000V range	+/-13 ppm (+/-7.75 ppm for 10 min )

[Theboel]

Hi TiN.
I really doubt about GPSDO or Rubidium but I think the best chance is OCXO with clean PSU.

[CalMachine]

CalMachine

I'm skeptical about "magnitudes" of improvements possible in any key performance aspects. Even for testing alone, hope you get that programmable JJA system in place, as that's pretty much only way to really split beans during testing whatever ideas we try. As was already mentioned, even if we design and get 9.5 digits, they mean nothing (other words - just noise) without linearity and stability to backup. And testing sub-ppm linearity is already hard enough, even with PJVS.

Having that said, replacement of onboard oscillator with more stable clock module is one of the things we can try in few months, once I confirm 3458B is fully stable and functional in stock condition. I'd expect rubidium-based source might be a better option with less phase-noise compared to GPSDO+OXCO combo. I have both anyway.

Btw, so far new ADC is good and stable, -0.01ppm drift over 7 days per SN18 procedure.

I also did a side test for 10V, 100V, 1000V ranges using same 10V reference.
To confirm the numbers, I ran same test with my reference 3458A using 10V out of 3245A (+/-0.5 ppm stability).

3458B 10V range Deviation	3458A test	3458A 24h spec
+0.25ppm, Reference, 10V range	0.0 ppm 9.999978 VDC reference from 3245A	+/-0.275 ppm
-1.6ppm 100V range	-1.5 ppm Datalog 100V range	+/-14 ppm (+/-3 ppm for 10 min )
+6.0ppm 1000V range	+3.4 ppm Datalog 1000V range	+/-13 ppm (+/-7.75 ppm for 10 min )

Yeah I doubt 'Magnitudes' as well, but it's still a hope!  The only reason why I popped out improvements of that order was the fact that DMM oscillators are like 0.01% at worst to 1ppm at best.  With a GPS referenced DOXCO or Cesium you can get stability of 1x10^-11 to x10^-12.  That is several orders of magnitude increase in accuracy of the oscillator for timing measurements.  I have high hopes that it could perhaps yield 1 order of magnitude increased measurement accuracy.  But yes, it is super high hopes, and most likely implausible.

Your numbers are looking really good!   

As Kleinstein said, there are still non-linearities and parasitic effects that would wash out any improvements made.

[ap]

Again, what is relevant here is the phase noise/jitter, not the absolute accuracy. The linearity of the A/D converter is already that good that it was hard to even determine it. In reality, it is probably much better. How would anybody identify an improvement after a mod? It will be very hard even with a JJA.
Linearity, btw, is not only driven by the A/D converter, but also by the amp stages in the input board.
Not sure overall it is a good idea to invest time trying to improve the world's most linear A/D converter installed in any test instrument.

[TiN]

Clock stability is not required to be much more stable longer than integration cycle, so all that effort would likely be just a snipe hunt. It's easy to confirm these claims however, much easier than anything else.

I'd agree that improvement on primary reference is a main area, also current source and current shunt resistors. Used resistors are not stellar by any means, leaving room for "easy" improvement with modern BMF's. If we find anyone with 3458HFL, we could verify this as well, since HFL's A1 seem to have better resistors as well, to support their improved resistance ranges accuracy. Other option to add active guard drive port for low-current measurement functions. Binding posts not very great for this application.

And get rid of that fan blowing over A1 PCBA with some thermal managment care. However there could be correction coefficients factored in 3458A's firmware for this effects compensation, so lots of hours testing is a must anyway.

[TiN]

3458B is getting final calibration love. Still has busted front terminals, so all calibration will be done using rear terminals.
But for sake of completeness, zero cal also ran on front. Innovation was required to attach a short, lol.



Don't try this at home, folks.. 

[Dr. Frank]

3458B is getting final calibration love. Still has busted front terminals, so all calibration will be done using rear terminals.
But for sake of completeness, zero cal also ran on front. Innovation was required to attach a short, lol.

Don't try this at home, folks.. 

Why didn't you exchange front and rear jacks?
I thought, that they are identical?

Frank

[Dr. Frank]

Yeah I doubt 'Magnitudes' as well, but it's still a hope!  The only reason why I popped out improvements of that order was the fact that DMM oscillators are like 0.01% at worst to 1ppm at best.  With a GPS referenced DOXCO or Cesium you can get stability of 1x10^-11 to x10^-12.  That is several orders of magnitude increase in accuracy of the oscillator for timing measurements.  I have high hopes that it could perhaps yield 1 order of magnitude increased measurement accuracy.  But yes, it is super high hopes, and most likely implausible.

Your numbers are looking really good!   

As Kleinstein said, there are still non-linearities and parasitic effects that would wash out any improvements made.

Hello,

A GPSDO is not a good idea, as the short time noise is bad, due to the satellites signal jitter.
A compact and decent OCXO is a much better idea, but look for good phase noise performance.
One of the metrological institutes, either NIST or PTB, they also modified the clock for some purpose, but I don't remember the background.

The LTZ1000A reference is quite noisy, about 1..2uV or 0.2ppm, but has good mid to long term stability.

Have a look on the different Allan diagrams, where best performance, about 0.02ppm or 8 digits, is achieved at about 10..20sec averaging.

At NPLC 1..100 you definitely see the short term noise, which might be improved by a less noisy reference, but for the price of bad long term stability.

If you would really achieve 9 digit performance at short to mid term time, I have my doubts, but an experiment would be great.

Frank

[TiN]

Yes, terminal blocks are same, but there is no need for that swap. This unit destined for lot of adventures, including (but not limited to) use of custom terminals block with active ohm guard and remotely programmable reversing.

Calibration today is to serve two purposes:
1. Use meter as one more transfer venicle between homelabs, so I can again test my accuracy on DCV/Ohm. This time also for AC, as SCAL to be done as well.
2. Get idea of aging rate and stability of used F5700 artefact cal using 3458 and guardband. Three artifact calibrations were done in sept/16, november/16 and today. I'll post details in related thread.

I found two last PCBs for my LTZ modules. If I get worty resistors I'd be able to have bank of 5 modules to try as A9 sub for lower noise. .

[CalMachine]

Yes, terminal blocks are same, but there is no need for that swap. This unit destined for lot of adventures, including (but not limited to) use of custom terminals block with active ohm guard and remotely programmable reversing.

Calibration today is to serve two purposes:
1. Use meter as one more transfer venicle between homelabs, so I can again test my accuracy on DCV/Ohm. This time also for AC, as SCAL to be done as well.
2. Get idea of aging rate and stability of used F5700 artefact cal using 3458 and guardband. Three artifact calibrations were done in sept/16, november/16 and today. I'll post details in related thread.

I found two last PCBs for my LTZ modules. If I get worty resistors I'd be able to have bank of 5 modules to try as A9 sub for lower noise. .

I'm interested to see the results of your mods to your 3458A.  Having a bank of 5 LTZ modules would be sexy!

[TiN]

I heard some 30kg of boxes have finally arrived. Wonder what that could be... Pile of shattered electronics or winner-winner-chicken-dinner? 

[Theboel]

I heard some 30kg of boxes have finally arrived. Wonder what that could be... Pile of shattered electronics or winner-winner-chicken-dinner? 

what 
2 pcs of broken 3458A ? 

[TiN]

No, but I do have extra broken A3 board. There were more parts in boxes than just 3458A for some other fun happy projects too.
And since I have meter in lab for these couple of days, let's hook it to 2000VA programmable AC source and test with difference mains, eh?

DUT is LTZ-reference, which is HP A9 STD in this case in a box, powered by +15V linear Agilent E3649 PSU.
It's calibrated value is +7.1846680 VDC +/-0.2ppm with unmeasureable tempco, obtained August 2016. Reference was sitting cold on the shelve since then.



Testing results are indeed interesting.



There seem to be visible relation to input mains voltage in sub-ppm level. I will also test with 200 and 240VAC, as well with 50 and 60, and perhaps 400 Hz settings.
Does this mean we need to monitor not just temperature/humidity but also mains power input for most accurate sub-ppm transfers? 

Btw, does anyone know how to remove binding post terminal without breaking stuff? I mean the rotating part from the threaded stud.

[Kleinstein]

Changing the frequency from 60 Hz to 50 Hz might need a call to ACAL, as this changes the timing and might also cause small gain changes.

With the AC source there is also the question about EMI - this could be an issue to the reference and the meter itself.

For the changes in the voltage, this can change the temperature distribution in the case, as the power dissipation at the voltage regulators will change and so will temperature gradients. So there is more than just one internal temperature.

[TiN]

There are ACALs after each voltage change and also on each 0.3c TEMP? change.

Only meter is powered from AC source, DUT reference PSU is constantly powered from regular mains.

As of temperature variance, that was exactly my thinking how different mains could make such result. Higher voltage on same transformer winding cause bit higher output voltages on secondary side, which in turn make LDOs run bit hotter. Since meter's own DCV path is about 0.1-0.3ppm/K it doesn't take much of the raise to bump similar change in measurement value. This just highlights another source of error for such accurate measurements to consider. If meter used in place where mains stability vary a lot, this could generate extra errors into otherwise stable readings. Perhaps double-transformer arrangement, such as used in Datron 1271/1281/4920/4950 and Fluke 8508A also helps to reduce this, as analog power section is powered there from intermediate regulator, so there is more decoupling from mains variations.

Another point for me is that meter's power well within 30W window, which is well manageable amount of power conversion need if we decide to try battery powered opeation. If we drop front panel and it's VFD, this number likely go down 5-10W. But I'd expect battery DC power may cause more bad than good, since 50/60/400Hz mains frequency used for syncronize/clock some parts of meter/ADC and so on. Yet still interesting to attempt.

P.S. Playing with different ideas.... no, these are not the ones which will go on meter

[Kleinstein]

With the meter running from a different supply frequency than the grid (and thus externally coupled "noise"), there could be a beat frequency signal superimposed to the signal. Worst case this could upset the ACAL adjustment and thus add some more or less random noise at that step. If could be interesting too look at some of the data where the beat frequency is low enough to be visible - might need an extra measurement.

The separate (higher frequency) transformer would make a battery powered operation easier. Still some of the meters still synchronize that frequency with the grid, to improve suppression by the integration time. It should at least keep the loss rather constant for the analog part.
I don't think battery operation for the 3458 is such a good idea - rather difficult at least, as there are several voltages in the signal referenced part.

A modified thermal layout might be possible, e.g. more separation of the voltage regulators, more thermal coupling. Are there thermal images of the 3458 internals ? Could be a nice reference for fault finding anyway.

[TiN]

Not as many voltages on both inguard and outguard actually, just +/-15 and +5s and 20V. Remove VFD and it's high-voltage DC-DC removes need for AC as well. Perhaps few battery packs from notebook could provide enough margin to downscale voltages for both negative and positive rails and have enough juice to run meter for a day or two. Custom regulator board would have a good spot in original A4 location.

Mains frequency impact can be tested by adding an clock input, locked to mains frequency, so both cases can be tested. So far it's running 50Hz and reading noise is in 0.15ppm range, within normal for LTZ ref.

I have thermal images somewhere, but never posted them before for some reason I don't remember now. Perhaps I'll take few more images on this unit this Friday after mains test done and before hauling it back to my homelab. It also missing one of inguard shields (ebay seller had it missing) so that could affect things a little too.

Another reason for custom A4 is to have more efficient regulator, so possibility of removing fan could be evaluated. The way airflow is directed with covers and placement suggest main purpose of the fan cooling A4 regulators and directing airflow onto A6/A5. So lot of things to play with 

Anyway, first I need to get meter back home and compare it's readings with my 2002s/3458A and all standards, so I can confirm their accuracy and not worry about calibration. As the moment I change fan, replace caps and do other things, calibration we done two weeks ago would be gonsky. I'll post report in next few days with magic numbers  .

[ap]

The 3458A sets the sampling synchronous to the mains frequency. A RESET e.g. does that (except for firmware 6 and below, a bug always sets it to 60Hz). It can be manually set, by LFREQ, in any case it is important that it is set correctly.

[TiN]

Here are the money shots. Few hot suckers detected



More to come...

[Kleinstein]

I don't see anything really bad on the thermal images. The power supply board is still not that hot - just a little. Also no surprise to see higher temperatures one the AC board: lots of fast amplifiers here an thus the high power. Still only a few parts with more than 60 C.

Some of the "diodes" near the reference seem to be rather hot and provide a hot spot in a possibly sensitive area. One has to keep an eye on these when using a different, higher power ref module.

The fast comparator at teh ADC board is rather hot - but also not on the main board and thus not a big issue. I don't think drift of this comparator would be important.

[TiN]

3458B in the house...finally, after 3 months of travel, remote calibration, ebaying and replacing boards, unit from this thread joined the pack.

Stashed it on top of poor 2002's, powered on, connected LTZ reference have currently online (I keep them all stored "cold", only power on for a week one at a time to do measurement), and here's what we see after 20 minutes runtime...



For those who don't play with ppm-calculator every day, here's the distilled data from photo:

LTZ module recorded value : +7.13665729 VDC, reference value obtained from comparison to F732B calibrated within 10 days to +/-0.55ppm in August 2016.
Bottom 3458A, this is my original rusty meter, never left homelab, DIY calibration transferred from K2002s from 7VDC cal. Reading is +7.1366591 V, +0.254 ppm
K2002 (GPIB 4) - handcarried and calibrated in August 2016. Reading is +7.1366605 V, +0.450 ppm
K2002 (GPIB 6) - handcarried and calibrated in August 2016. Reading is +7.1366588 V, +0.212 ppm
Top 3458A - is this unit from thread. Reading is +7.1366579 V, +0.086 ppm

Can I say now that I've joined sub-ppm 10DCV accuracy club? 

[ManateeMafia]

Looks great. I am glad it all came together for you in time for the Chinese New Year.

Btw,  I see no problem in having too many clocks.

[VintageNut]

You are the honorary president of the club. From a repair and restoration standpoint, as well as volume of work, you are supreme.

I very much enjoy reading your methods and results.

Can I say now that I've joined sub-ppm 10DCV accuracy club?