Keysight 34470A, Calibration and ACAL problems

[HighVoltage]

I have two Keysight 34470A and one 34465A.

The 34465A works perfectly and is very stable over a long time with no noticable drift and the ACAL function works perfectly as well.

But the two 34470A have shown huge drifts over the years and I am wondering, if this is caused by using the ACAL function.
Because of these drifts, Keysight exchanged both instruments and delivered two brand new instruments to me with new calibration certificates about 18 month ago.

At present time, I have a very stable and verified 10.000,040 V reference as a loaner and use it to calibrate all my 6 1/2 digit meters. All my 6 1/2 digit meters were within +/- 40 uV of 10.000,040 V before calibration, except the two 34470A. Even most of my 34401A were between +/- 20 uV!

The two 34470A showed this:
34470A_#1 = 10.000,135 V (+95 uV)
34470A_#2 = 10.000,178 V (+138 uV)

After calibration against the reference, the 34465A and the two 34470A showed a perfect value 0f 10.000,04x V
34465A       = 10.000,04 V
34470A_#1 = 10.000,040 V
34470A_#2 = 10.000,042 V
And within 1 hour after the calibration there was no significant change.

Then about 1 h later I pressed ACAL on all three DMM and got the following results:
34465A       = 10.000,04 V (no change)
34470A_#1 =   9.999,965 V (-75 uV)
34470A_#2 = 10.000,178 V (+138 uV)
And immediately after that, I can see a drift in in trend chart.

Has anyone here experience the same with their 34470A?
I know of at least one eevblog member who has the same problem on his 34470A and would rather return it to Keysight for a refund.

I have followed the calibration procedure of the service manual and it passes the calibration perfectly.
May be one should NOT use the ACAL function of the DMM?

Any ideas?

[Magnificent Bastard]

Wow!  There appears to be something really wrong with their ACAL design and/or firmware technique.  The LTZ000A references in these meters appear to be a design where less care was taken (as opposed to the reference in a 3458A) to minimize thermal issues.  I particularly dislike the fact that they placed it right in the air stream of the fan; and also their choice of using a cheap header for connection to the main board; these are sources of drift against temperature; and having a "breeze" flowing across this cheap connector (under the reference board) cannot be helping matters.

In a DMM that relies on ACAL to stay accurate, it is imperative that the voltage and resistance (artifact) standards be rock-solid over time and temperature.  This does not appear to be the case at least with the voltage artifact reference.  You did not make any tests of the resistance readings through ACAL cycles; but I suspect that they are not having the same issue as the voltage readings.

It would be an interesting experiment to attach a cable to the 7V reference on the main board.  Then put the meter in a chamber, and plot what happens to this voltage as the ambient temperature is varied.  This voltage must be rock stable, no matter what.  Perhaps a second cable can be attached directly to the 7V reference module; this would tell you if the voltage variance is due to the reference itself, or the cheap 0.100-inch header used to connect the 7V reference output to the main board.

[Kleinstein]

Even a rather poor thermal design can hardly explain some 100 µV of drift in the 10 V range, thus some 70 µV at the 7 V reference.

It is odd to come back to nearly the same + 138 µV on the meter number 2
This looks like something is really wrong in the ACAL procedure.

I simple test without opening the case would be a repeated (e.g. 5 times) ACAL call and see how much the voltage reading or maybe cal constants read back from the meter develop with repeated calls. Ideally there should be just a very small amount of noise (e.g. less than 0.1 ppm).

[iMo]

https://community.keysight.com/thread/23709
It could be with the new FW they introduced a bug into the 70A..

[HighVoltage]

https://community.keysight.com/thread/23709
It could be with the new FW they introduced a bug into the 70A..

Yes, this was a known problem on earlier FW but solved since FW v2.14
FW v2.14 and higher solved the problem completely on the 34465A as far as I know.

I also believe that by following the Service-Manual calibration procedure, I am doing it all correctly.
And the 34465A works perfectly and therefore shows that the CAL procedure is correct.

Also, both 34470A were factory calibrated in 2017 and the calibration was not touched until today.

Measuring the 7V of the LTZ1000 output is a good idea and see the relationship after an ACAL.
I will try that.

[Magnificent Bastard]

Even a rather poor thermal design can hardly explain some 100 µV of drift in the 10 V range, thus some 70 µV at the 7 V reference.

It is odd to come back to nearly the same + 138 µV on the meter number 2
This looks like something is really wrong in the ACAL procedure.

I simple test without opening the case would be a repeated (e.g. 5 times) ACAL call and see how much the voltage reading or maybe cal constants read back from the meter develop with repeated calls. Ideally there should be just a very small amount of noise (e.g. less than 0.1 ppm).

I totally agree with you on all of your points.  I would also like to see the results of the test I outlined.  What we need are more clues, so we can narrow down what might be going wrong.  It could be firmware, or hardware, so some nasty combination of the two that is causing this.  Keysight will be better able to address the issue if they know exactly what it is; (although I have to believe at this point, they "have their best guy on it right now").

[e61_phil]

I did a short test with a 34470A here. I run ACAL and the 34470A showed +4µV after 55min ist showed -5µV. That isn't as much as you see, but significant. I love ACAL meters

[Dr. Frank]

Hello HighVoltage and e61_phil,

I remember to have characterized the first '470A of HV: https://www.eevblog.com/forum/testgear/keysight_s-new-34465a-(6-5-digit)-and-34470a-(7-5-digit)-bench-multimeters/msg889215/#msg889215
Back then, I did not see that drift at all, even after ACAL. Only the LTZ had drifted -5ppm after 10 months of use, but that's to be expected and fully inside specification.
As the 465A and the '470A share exactly the same motherboard and firmware, I really can't imagine, how ACAL should influence the LTZ1000A reference, especially this extreme drifting.
There's a known common behavior of both versions, that after an ACAL, a slight decrease of about 1..2ppm of DCV reading over 1h can be observed...somebody wanted to clarify that with KS, but I did not yet observe any further feedback.

As this 10.000 040 V reference is probably that special reference loaned from another volt-nut from Bremen, who also REALLY dislikes all ACAL featured instruments   , I suppose that today, on 2nd of April, your '470A might just behave much better? 
Anyhow, how would one 'calibrate' a '365A or '470A from just one 10V reference? These are no Artifact Calibration instruments, like the 3458A.. Right, Phil?

Anyhow, if that's NOT been an April fool's, I'm really curious, what KS might say about that design flaw.. and we still have not completely reverse engineered the '470As LTZ1000A circuit and its connection to the main PCB.. for an upgrade solution...

April greetings to both of you!
Frank

[HighVoltage]

Hello Frank

LOL!
No, may be this was bad timing to post it on April 1st.
But this is a serious matter with both of my 34470A and it has not vanished over midnight....

To calibrate the 10V range of the 34470A I first calibrate the zero reference with a short block and store the data.
Then calibrate +10V with the +10.000,040 V
Then calibrate -10V with the -10.000,040 V (reversed polarity)
Then store the data.

After the ACAL yesterday, the instrument stabilized within a few hours and I made another 10V calibration.
This is now running almost 20h and seems to be very stable.
I will do another ACAL shortly to document the difference.

Keep in mind, both instruments had a factory calibration from 2017 and were not touched in the calibration
since, except ACAL.
And in these 18 month, they drifted (+95 uV) and (+138 uV)
For me that is too much for an instrument used in metrology.

In comparison, my Keithley DMM7510 read from the beginning + 3ppm high (probably because of sloppy calibration) but has not moved at all in several years.

[TiN]

While I do agree that 34470A shouldn't change THAT much, it is not really metrology targeted instrument. 

Also there is bit of terminology conflict in your post, HV.

Then calibrate +10V with the +10.000,040 V
Then calibrate -10V with the -10.000,040 V (reversed polarity)
Then store the data.

That is not calibration, but adjustment, and it contradicts

had a factory calibration from 2017 and were not touched in the calibration
since, except ACAL.

Unless you had reverted original calibration ROM to prior state after adjustment? 
Calibration is measurement comparison only, not the adjustment. Isn't ACAL on 34470A only internal correction mechanism, that does not need external reference?

On a side note, since your reference source type is guessed by Dr.Frank (correctly?) I observed some wierd behaviour output with Wavetek/Fluke 7000. Reliable results were obtained only when I had this ref on highly-isolated power source/battery. Usual brick that comes together with F7000 isn't a good source for ppm-accurate comparisons. Other voltnuts had questionable results due to noisy DC/DC in W7000 as well. So ideally such a test should be performed with multiple stable sources for sanity check. I'm still working on schematics of my W7000 

Keep in mind, I don't have/never touched 34461/65/70A before, so I might as well be wrong here.
 
Dr.Frank  For whatever Keysight do on 3459A, I hope they don't break 3458A's ACAL  .

[Dr. Frank]

Hi HighVoltage,
ok, I assume that you still let your '470As run 24/7.
Therefore, they are inside specification, for 10V dc, i.e. 16ppm / 1yr., 20ppm / 2yrs., and you've seen 10 and 14ppm after 1 1/2 years.
If you subtract these 8ppm of 'basic accuracy' for 24h, you get about these 8ppm/yr. and 14ppm/2yr. specs of the 3458A, caused by the same violated LTZ1000A @ 95°C (+).
Metrology grade instrument? Nope, neither is the 470A, nor the 3458A, only because of that LTZ reference.

And because the 470A  has the same 6 1/2 hardware (these 8ppm..)  on the main board, and no real artifact / ACAL calibration to compensate for that, it's a good and bit more stable 34401A replacement, but nothing more.

If you have real 'guts', (.. Olli Kahn ...), then you may pimp your LTZ references to 12.5k/1k or 65°C, and send them off for re-calibration and adjustment, and live happily with < 2ppm/sqrt(yr) ever after.


Hi TiN,

I doubt that KS will ever design a successor.. the 470A is good enough for that industrial market, at least on paper.

Also, all of the old eggheads left the company, or pitiful passed away..

But if they do, they really should keep the existing ACAL procedure, but add Vishay BMF resistors (or something like FLUKEs hermetic TF arrays) all over the place, like inside the Fluke 8508/58/88, AND should consult us for a much better LTZ reference. (Daniel B., did you hear the prophecy?)

Such a model update would also silence this certain ACAL denier 

Frank

[HighVoltage]

Dr. Frank
You guessed correctly on the 7000 and it seems to be very stable.

Good points on the yearly drifts of the Keysight  instruments.
May be I am just surprised that some of my 34401A seem to be more stable.
And yes, my two 34470A and the 34465A are running 24/7

Well, I have a very good 10V baseline now on all my DMMs.
It looks impressive, when all of them show the same number on the display to the last digit.

Yes, point taken on not being metrology gear.

[HighVoltage]

Ok, next test.
Both Instruments were running very stable almost a full day.

Reference voltage: 10.000,040 V

And the instruments show:
34465A       = 10.000,04 V
34470A_#1 = 10.000,036 V (-4 uV)
34470A_#2 = 10.000,035 V (-5 uV)

Then next, I performed ACAL 5 times with 5 min resting in between.

And the instruments show:
34465A       = 10.000,05 V
34470A_#1 =  9.999,969 V  (- 67 uV)
34470A_#2 = 10.000,178 V (+ 143 uV)
 
In comparison to the 34470A, the 34465A is rock solid.

 

[Magnificent Bastard]

So, this is a REAL PROBLEM--- and not some imaginary thing nor is it an April Fool's joke.  Furthermore, this is NOT FUNNY Keysight !!!!

@Keysight DanielBogdanoff:

   We, the "early adopters" of the (rather expensive) 34470A deserve some answers.

[Keysight DanielBogdanoff]

Hey you Magnificent Bastard (I've never called anyone that until now!), I'll dig into it on my end.

Jokes aside, have you contacted your local support center? For this type of thing, they have access to a couple issue escalation avenues that I don't get to use. Most of my support here is unfortunately ad hoc.

I will dig in, but I highly recommend you also give our support folks a call. They're very good.

[HighVoltage]

Hello Daniel,

Thanks for the response, I have sent you a PM.

[HighVoltage]

Just to show, how stable the Keysight 34461A is.
I had them turned off over night and turned them back on this morning.
And they move very beautifully towards 10V. Interestingly, one does it differently than the other 3.

This new reference source is a Fluke 731B at 10.000,01 V

[HighVoltage]

Last night, before I went home, I had the two 34470A adjusted again on the 10V range

This new reference source is a Fluke 731B at 10.000,01 V
And both instruments stayed stable over night.
(NO ACAL !)

[Andreas]

The two 34470A showed this:
34470A_#1 = 10.000,135 V (+95 uV)
34470A_#2 = 10.000,178 V (+138 uV)

After calibration against the reference, the 34465A and the two 34470A showed a perfect value 0f 10.000,04x V
34465A       = 10.000,04 V
34470A_#1 = 10.000,040 V
34470A_#2 = 10.000,042 V
And within 1 hour after the calibration there was no significant change.

Then about 1 h later I pressed ACAL on all three DMM and got the following results:
34465A       = 10.000,04 V (no change)
34470A_#1 =   9.999,965 V (-75 uV)
34470A_#2 = 10.000,178 V (+138 uV)
And immediately after that, I can see a drift in in trend chart.

Any ideas?

Hmm,

the behaviour of #2 reminds me of some roumor regarding personal weight scales:
to hide the drift/instability of the scale they remember the last showed value(s).

If a value of similar weight (with small difference) as one of the last values is measured the display is (more or less) the according last stored value.

with best regards

Andreas

[Towger]

Any ideas?

If a value of similar weight (with small difference) as one of the last values is measured the display is (more or less) the according last stored value.

If you can't meet the specs/requirements, fix the firmware.  A common enough occurrence these days. Often back fires in big ways as VW and Boeing have discovered.

[Keysight DanielBogdanoff]

Hi Everyone,

I met with the tech support manager today to get an update on where they are with this. The division support folks are working on verifying/replicating this on their 34470As and will proceed from there.

One thing that would be helpful is, if you are experiencing this, please send me your serial number! DM is fine, or you can email me directly at daniel.bogdanoff@keysight.com.

Thank you!

[eplpwr]

Just wanted to chime in. I'm seeing a jump (positive offset) after ACAL, then slowly returning to "baseline" after 5-10 minutes.

Normal monitoring of 732A voltage reference is essentially a straight line, after ACAL it will jump and then over time return to previous value. The behaviour is 100% reproducible.

[Magnificent Bastard]

Just wanted to chime in. I'm seeing a jump (positive offset) after ACAL, then slowly returning to "baseline" after 5-10 minutes.

Normal monitoring of 732A voltage reference is essentially a straight line, after ACAL it will jump and then over time return to previous value. The behavior is 100% reproducible.

I have this too.  I always wondered why that was happening-- I assumed it is something heating up (or a capacitor charging) somehow during the auto-cal process, and then settling down.

What firmware version is your DMM on?

[Kleinstein]

The time constant looks a little like a thermal effect. It is rather slow for a real RC and I won't expect that much dielectric absorption. With a Keithley meter I would suspect something like averaging on zero readings - but this a Keysight and should not have that bug included.

Just for a test, how does it look if one has the stable reference connected, change the range to 100 V or 1000 V for some 5 minutes and than back to 10 V. I would not be so much surprised to see a similar settling. Still it would be a bit disappointing to see so much thermal effect.

[eplpwr]

I have this too.  I always wondered why that was happening-- I assumed it is something heating up (or a capacitor charging) somehow during the auto-cal process, and then settling down.

What firmware version is your DMM on?

I'm on the latest version - 3.0. I didn't see this "jump" happening when the meter was on the previous 2.x version, at least it wasn't as pronounced. Since the measured value gets back on track rather quickly I can live with this. I only seen this "bump" on the '470; I have an 34465A with s/w v3.0 as well, and if I measure the same stable voltage source the '465 is ruler flat. The '465, however, actually needs ACAL if the temp has changed, whereas the '470 seems to make such minor adjustments that not even the least significant digit changes.

[Grandchuck]

Just wanted to chime in. I'm seeing a jump (positive offset) after ACAL, then slowly returning to "baseline" after 5-10 minutes.

Normal monitoring of 732A voltage reference is essentially a straight line, after ACAL it will jump and then over time return to previous value. The behaviour is 100% reproducible.

See a similar behavior with a 34465A.

[eplpwr]

The time constant looks a little like a thermal effect. It is rather slow for a real RC and I won't expect that much dielectric absorption. With a Keithley meter I would suspect something like averaging on zero readings - but this a Keysight and should not have that bug included.

Just for a test, how does it look if one has the stable reference connected, change the range to 100 V or 1000 V for some 5 minutes and than back to 10 V. I would not be so much surprised to see a similar settling. Still it would be a bit disappointing to see so much thermal effect.

@Kleinstein

I have tested to change to 100V range and let the meter sit there for >10 minutes, then changed back to 10V range. I can not see any "jumping" in this case. Since the resolution is 1/10th in the 100V range, and the input Z is changed to 10M, the measurements are obviously a bit off. I think it's quite visible when I change to the 10V range in the picture below.

[HighVoltage]

Kleinstein

Here I switched the 34465A and the 34470A to manual 100V range, waited 5 min and switched back to 10V.
The 34465A seems to be more stable
The 34470A shows a small jump.

In comparison I also added the Keithley DMM7510 with am much larger jump.

Test's done on a stable 10.000,040 source.


 

[HighVoltage]

I have an 34465A with s/w v3.0 as well, and if I measure the same stable voltage source the '465 is ruler flat.

After pressing "Autoscale Once" the best resolution, "span", the 34465A does is 100 uV but the 34470A does 10 uV.
So, to really compare correctly, one must set the vertical scale manually also to 10 uV span.

But I agree, my 34465A is also "ruler-flat" even when the vertical scale is manually adjusted.

The '465, however, actually needs ACAL if the temp has changed, whereas the '470 seems to make such minor adjustments that not even the least significant digit changes.

Interestingly, my 34465A almost is not changing after an ACAL and stays rock solid.
And it has been like this since I have it.

[HighVoltage]

In comparison, the Keithley DMM7510 is also rock solid after an ACAL
This test was done on a 10.000,010 V source.
My DMM7510 has a +30uV offset since I have it (Sloppy Keithley Calibration, I guess)
But otherwise it is not drifting at all, that I have noticed.

[Kleinstein]

The time constant looks a little like a thermal effect. It is rather slow for a real RC and I won't expect that much dielectric absorption. With a Keithley meter I would suspect something like averaging on zero readings - but this a Keysight and should not have that bug included.

Just for a test, how does it look if one has the stable reference connected, change the range to 100 V or 1000 V for some 5 minutes and than back to 10 V. I would not be so much surprised to see a similar settling. Still it would be a bit disappointing to see so much thermal effect.

@Kleinstein

I have tested to change to 100V range and let the meter sit there for >10 minutes, then changed back to 10V range. I can not see any "jumping" in this case. Since the resolution is 1/10th in the 100V range, and the input Z is changed to 10M, the measurements are obviously a bit off. I think it's quite visible when I change to the 10V range in the picture below.

It's obvious the 100 V range readings are not good. The interesting part is the recovery in the 10 V range. It looks like there is essentially no extra recovery here. This indicates no much thermal effect with the ADC just from a different reading.  So more like an ACAL specific effect.

A possible reason I could think about of could be a numerical temperature correction that used temperature reading from the last 5 minutes or so. During ACAL those temperature readings may be missing or off for some reason. There is a good indication the Keithlay 7510 uses such temperature corrections - this helps to get good readings even shortly  after power on. So maybe the 34470 uses some similar corrections too to get better stability than the 34465.

If the effect is recovering after something like 5 minutes I would not worry so much about this. However it would be nice to know for sure.

[eplpwr]

Interestingly, my 34465A almost is not changing after an ACAL and stays rock solid.
And it has been like this since I have it.

Well, it's a matter of both selection and also statistics/distribution - you obviously got the "golden" LM399AH of the lot! 

[kj7e]

Interestingly, my 34465A almost is not changing after an ACAL and stays rock solid.
And it has been like this since I have it.

Well, it's a matter of both selection and also statistics/distribution - you obviously got the "golden" LM399AH of the lot! 

My 34465A is also surprisingly stable.  I do see a slight (like 2-3 ppm) downward drift over about 15 min after ACAL, but it seems to land on the right readings after the drift.  If the room temp changes more than a few deg, the ACAL does make an improvement. 

[Magnificent Bastard]

I have this too.  I always wondered why that was happening-- I assumed it is something heating up (or a capacitor charging) somehow during the auto-cal process, and then settling down.

What firmware version is your DMM on?

I'm on the latest version - 3.0. I didn't see this "jump" happening when the meter was on the previous 2.x version, at least it wasn't as pronounced. Since the measured value gets back on track rather quickly I can live with this. I only seen this "bump" on the '470; I have an 34465A with s/w v3.0 as well, and if I measure the same stable voltage source the '465 is ruler flat. The '465, however, actually needs ACAL if the temp has changed, whereas the '470 seems to make such minor adjustments that not even the least significant digit changes.

Interesting!  I wonder if it's possible to "downgrade" the firmware to the last 2.x version, and see if it still happens.  (I would ask Keysight before you do this-- it might "brick" the meter to go backwards in firmware versions).

If the problem goes away at 2.x and then reappears at 3.x, then that would be a huge clue for the Keysight people!  I'm certain that they have source code control, and they will be able to see exactly what was changed in the ACAL from 2.x to 3.x (and who did it too)...

[Keysight DanielBogdanoff]

Quick update, the product support teams have been characterizing a couple units since late last week when this popped up. So far we do see some offset induced by ACAL, but not as big as described here and not anything out of spec.

We have been and will continue to working with the individuals seeing this over e-mail, we may reach out to swap/borrow some to attempt to recreate the problem.

[maxwell3e10]

Here are some very quick data not from a calibration lab! All meters have not been calibrated since manufacture. Voltage source is a REF102.

KS34470:   9.996340 before autocal (previous autocal maybe performed 1 year ago)
                 9.997021 after autocal
KS34465:   9.99689 before autocal  (previous autocal maybe performed 1 year ago)
                 9.99685 after autocal
DMM7510: 9.996920 (didn't do autocal, it takes too long)
HP3457:    9.99690

So, Keysight 34470 does move a lot with autocal, but maybe closer to the right value in this case. But the bottom line, why bother with all these new instruments if a 30 years old uncalibrated HP3457 is just as accurate? 

[iMo]

Quick update, the product support teams have been characterizing a couple units since late last week when this popped up. So far we do see some offset induced by ACAL, but not as big as described here and not anything out of spec.

We have been and will continue to working with the individuals seeing this over e-mail, we may reach out to swap/borrow some to attempt to recreate the problem.

Has the fix been found?

[HighVoltage]

Keysight is working on this problem now and really have put their attention to it.
Thank you Keysight !!!

[iMo]

And the result? What was the issue?

[HighVoltage]

Keysight is still working on this and it might take a little longer.
So it seems we have to have some patience.

[jaromir]

Right now I'm going to setup new electronics laboratory, deciding what TE to buy and I'm expected to submit final list shortly. I thought of buying a few lower grade (lower than 34470A) multimeters and one 34470A to rule them all; but reading this thread made left me a bit unsure about it.

I wonder whether this is problem systematic to all 34470A units or is it just a few isolated cases?
In any case, does any of the affected users have (at least preliminary) fix to the problem?

[eplpwr]

In any case, does any of the affected users have (at least preliminary) fix to the problem?

My fix is to wait 3-5 minutes after ACAL before re-trusting the 34470A readings. Mine is well within spec BTW, so I trust it - checking it against two 3458A at least weekly. Also, I have DMM7510 and Fluke 732A for intercomparisons.

[e61_phil]

Our 34470As drift less than 1ppm after ACAL. That is only around 10% of the 24h spec.

[GEOelectronics]

34470A_#2 = 10.000,178 V (+138 uV)
And immediately after that, I can see a drift in in trend chart.

Has anyone here experience the same with their 34470A?
Any ideas?

I'm beginning to think so yes. I've had mine for a year or two now, I picked it because of the autocal feature and the LTZ-1000, and bought it expressly for a few specific measurements:

1)1000 V with 1G input impedance- Well it doesn't and it's my fault, only up to 10V can have 1G Ohm input, above that, only 10M.

2) Measure 1G Ohm resistors with high precision. It should but it doesn't really. Up to 500M it seems fine, at 1G it takes forever to decide, basically .

For example, I've been working with 99M 0.1% resistors. What I need is 990M Ohms with precision and stability for a HV divider. The 34470A reads 99M swiftly and with stability, all are within tolerance./ Any two read 2X 99M, any 3 read 3X 99M and any 5 read 495M just as they should. Try to read all 10 resistors it gets muddy, same with a single 2W 1G 1% resistor

I trust the resistors, they show good on every instrument I own. I trust the series stack of resistors, they read fine on an ESI Guarded Wheatstone bridge and on an HP 4329A.

Some of the difficulty is in my inexperience with the modern high end, high tech instruments.

So far the ESI has proven the best instrument for me, it gives enough digits to be meaningful, and is rock steady, is fast and accurate. The 4329A is good and can read at 1000V which is what I need, but is analog, no resolution and rated rather poorly, although it does work very well.

I'm not disappointed yet but am confused and confounded, that something that should be easy is so hard to master.

George Dowell

[TiN]

You'd have much better results if you try use external stable 1kV source applied to your 990M/10M divider and measure output voltage by 34470 on 10VDC range instead.
Megaohms+ is where leakage/cabling/shielding start to matter a lot.

[GEOelectronics]

Understood. Thanks. Using a NIM HV PSU (Nuclear Instrumentation Module), pretty stable. Made to drive large sensitive photomultiplier tubes (can run 100M Ohm dynode string, down to <100k Ohm)  where any HV change will ruin measurements.

1)Measure resistors with '4470A = 990M
2)Measure NIM HV PSU with '4470A = 1000.0VDC
3)Measure NIM with 990M resistor in series with '34470A on 10V = 10.07

It's as if the 10 M input is not exact, could this be? Will try it with 34470A set to 10V but 1G input and an external precision 10M resistor tomorrow.

It has to be well below 1% accuracy to be of any use to me.

George Dowell

[maxwell3e10]

Yes, the spec on 10M input impedance is +/- 1%

[GEOelectronics]

Yes, the spec on 10M input impedance is +/- 1%

OK. Well it's within specs then. Thanks.

George Dowell

[e61_phil]

I wouldn't use the internal 10Meg as part of the divider. I made bad experiencies with such a setup.
It is much better to switch the meter to high Z and have the whole divider external.

[Dr. Frank]

If you need really high impedance @ 1kV, best way would be a differential method, i.e. creating 1kV with a stable/precise DC source / calibrator, and then sensing the difference at high level, maybe also with the 34470A. That way, you'll also get high precision.
One these old Fluke 88x Differential Voltmeter really had a 1kV reference inside.

The other way is to use a dedicated HV divider, which usually is base on the 10M input impeidance inside the DVM.
You'll have to calibrate the divider ratio first at a lower DC level, and then maybe set the '470A for a math multiplier.
I wonder if there are still 990MOhm divider available..

Frank

[e61_phil]

The 34470A isn't able to float on 1kV (If I remeber correctly max. 500V). There are nullmeters which are able to float on 1kV. Most of the benchDMMs are not designed for that.

To use the internal 10Meg is also a bad idea for a divider, as I already said. Build a proper divider and measure it with high Z. Or build an active one. I did it for up to 10kV and it works also very well.

[GEOelectronics]

I wonder if there are still 990MOhm divider available..

Frank

Hi Dr. Frank, can you reference some (or one?) 990 Ohm divider even from the past?

I can't even find any more 99M 0.1% HV Hymegs to build them up.

Most of the "HV Probes" for DVMs are much smaller value, no good on a current limited bias supply.

I've had some special made in China for a project but not anything like 0.1% tolerance. Fortunately the project didn't require accuracy, just stability (in that one part).


George Dowell

[GEOelectronics]

The 34470A isn't able to float on 1kV (If I remeber correctly max. 500V). There are nullmeters which are able to float on 1kV. Most of the benchDMMs are not designed for that.

To use the internal 10Meg is also a bad idea for a divider, as I already said. Build a proper divider and measure it with high Z. Or build an active one. I did it for up to 10kV and it works also very well.

Please explain term "float", not understanding the reference, thanks

George Dowell

[e61_phil]

With float I meant you are not allowed to lift the LO input of the 34470A higher than 500V above ground (with a 1000V source for example). On the most bench DMMs only the HI input can go up to 1000V (or 1500V if you add both up). This is normally marked on the unit.


What kind of HV Probes for a DMM do you mean? The ones I know (1:1000) have an input resistance of 1GOhm.

Ohmcraft and Caddock also sell dividers with good TCR.

And again: Don't use the internal 10Meg. There is no specification about TC, VC and so on (for the meter only tracking is important together with the rest of the input circuit). I made a few tests with 90Meg in series to a 34401A to measure up to 10kV in the 1kV range and the readings where very unstable.
If you use high impedance dividers also take care about the input current of your meter.

[e61_phil]

I attached a picture of the divider box. This box was only a quick and dirty build to verify a HV measurement system.

It was build out of Caddock USF 20Meg resistors out of the drawer. Two in parallel and in the beginning nine of such sets in series to build 90Meg. That worked so badly, that I decided to put in another pair of 20Meg resistors to achieve 100Meg (one can see, that the uppermost Caddocks wasn't planned to be there). I added a Vishay 100k Z201 and some wire wound resistors out of the drawer to set the attenuation to 1:1000. The small 25ppm/K metallfilm resistor is just for fine tunning and doesn't affect the TC.
Overall TC of the box is ~1.5ppm/K

[GEOelectronics]

With float I meant you are not allowed to lift the LO input of the 34470A higher than 500V above ground (with a 1000V source for example). On the most bench DMMs only the HI input can go up to 1000V (or 1500V if you add both up). This is normally marked on the unit.


What kind of HV Probes for a DMM do you mean? The ones I know (1:1000) have an input resistance of 1GOhm.

Ohmcraft and Caddock also sell dividers with good TCR.

And again: Don't use the internal 10Meg. There is no specification about TC, VC and so on (for the meter only tracking is important together with the rest of the input circuit). I made a few tests with 90Meg in series to a 34401A to measure up to 10kV in the 1kV range and the readings where very unstable.
If you use high impedance dividers also take care about the input current of your meter.

1)"With float I meant you are not allowed to lift the LO input of the 34470A higher than 500V above ground" OK I understand. OK I understand, thank you. Fortunately all our measurements are referenced to "ground".  Quotes used because much of the generators are battery powered handheld, so no actual earth ground. Some photomultiplier tubes are run with negative HV to reduce noise, but this puts the mu-metal magnetic shield floating at kV levels referenced to ground. I encourage users to convert to +HV the so the mu-metal can be grounded.

2) "What kind of HV Probes for a DMM do you mean? The ones I know (1:1000) have an input resistance of 1GOhm." The ones I have been using for general purpose HV power supply servicing (not current limited bias supplies for nuclear measurements) are of usually add-on to a 3 1/2 or 4 1/2 digit handheld DVM's (or analog VOM or VTVM) regular probe and only one I know of is even close to G Ohm is one Fluke model.

3) "And again: Don't use the internal 10Meg. There is no specification about TC, VC and so on " I agree 100%. Doing so has become commonplace in some circles, worldwide, and I am gathering anecdotal references to refute the practice on several levels.

4) "I attached a picture of the divider box. This box was only a quick and dirty build to verify a HV measurement system." I like that thanks for showing, it looks very "industrial"., which is exactly what I do. The KS 34470A is my first toe dip into precision, after 50 years of being in the electronics service industry. My normal instrumentation is more like what you would see in a combination typical bio-med and nucleonics type service facility.  Our standards are in keV produced by man made radioisotopes.

5) OK it's time for me to get on with this and break out the silicone wire, PTFE wire, Kapton tape and get back to the workbench to refine this project based on all the good advice given by members of this forum.  Thank you all.

George Dowell

[GEOelectronics]

If I may, attached is a typical HV PSU that I make, I call it an HVG for High Voltage Generator.

This one uses 2 of the made-to-specification glaze 800M Ohm resistors, one controls and IC that is part of the HV set reference (200-2500V) while the other feeds the IC that is the onboard HV readout. (works with either analog meter or digital panel meter). Also the small transformer was custom made for me.

In use this is sealed inside a box where a safety resistor in the 1 to 10+ MegOhm range is in series with the output.

Simpler versions of the HVG have non adjustable, fixed HV output and no onboard HV meter driver, but still the safety resistor, so the service/calibration HV measurement must be done through the unknown value M Ohms safety resistor, therefore the need for 1G Ohm HVM (HV Meter)

Thanks for helping. me understand what I was doing wrong.

George Dowell

[GEOelectronics]

If you need really high impedance @ 1kV, best way would be a differential method, i.e. creating 1kV with a stable/precise DC source / calibrator, and then sensing the difference at high level, maybe also with the 34470A. That way, you'll also get high precision.

A HV Voltage comparator is the very next project- well maybe after the LTZ1000 kit build and the completion of the AD2702UD builds.....

George Dowell

[e61_phil]

A HV Voltage comparator is the very next project- well maybe after the LTZ1000 kit build and the completion of the AD2702UD builds.....

If you already have such supplies you could use a battery powered DMM. Perhaps in combination with your 34470A. A 1kV source with low output impedance will drive the input of the ground referenced 34470A and also the LO from a fully isolated batterie powered (low cost) DMM. With both readings you have a very accurate and high impedand measurement.
I did such crude measurements a few times to measure a couple of mV floating on several kV.

But it sounds you need a robust solution for a service technican.

[iMo]

Keysight is still working on this and it might take a little longer.
So it seems we have to have some patience.

Back to the topic of this thread - has KS fixed the issue with 34470A cal and ACAL already?

[HighVoltage]

No solution so far, Keysight is still working on this issue.

[iMo]

Any feedback from KS?

[HighVoltage]

Only feedback so far is, that Keysight is working on it.
I have no problem for this to take a little longer, as long as all the problems are solved at the end.

[iMo]

If I were KS I would offer you a free replacement for the 3458A/BE as the compensation for the 7 months without the gear.. 

[Andreas]

NO they will not.
Highvoltage could find the same problem in the 30 year old firmware too.   

with best regards

Andreas

[djac]

Any update? I'm thinking about getting a 34470a. But as long as this problem is not solved or cleared up, I will refrain from it.

Regards Dieter

[HighVoltage]

Please keep in mind that not all owners of a 34470A have reported the same ACAL problem as I have.
This should not hold you up from buying one of these great instruments.

[djac]

No, it's not that simple. I have the impression that the problem is being swept under the carpet in silence and not played with open cards.

Surely the jumps to ACAL are extreme with your two devices and a special case. But did Keysight replace the devices without any trouble and how do these new devices behave?

Because after all Keysight has admitted that this problem exists, albeit to a lesser extent. And some forum members have confirmed this in this thread.

But after the announcement here to solve the problem, you just don't hear anything anymore. This makes you feel insecure.

You can probably live with it if you have to wait 5 minutes after an ACAL until you can take measurements on the edge of accuracy, but that this is assured is not confirmed by Keysight.

So one wonders if there is any advantage to paying double the price of a 34465a that doesn't have this problem. Especially since the noise and linearity measurements presented here in the forum do not at all indicate that the 34470a really has a clear advantage here. Thus, it looks more as if the two devices do not differ in linearity and only slightly in noise.

So why a 34470a versus a 34465a?

Regards Dieter

[HighVoltage]

All I can say is that Keysight is working on this and I have confidence that it will be solved.

It highly depends on what you want to do with the instrument.

The 34465A has all features of the 34470A and is ultra stable not just in my lab but also reported by many others.
And if you use software to read out the data, you even get 7 1/2 digits on the 34465A.
And the 34465A is not effected by the ACAL drift at all, as far as I have noticed.

[iMo]

34465A (LM399 ref) cannot be as "ultra stable" as the 34470A (LTZ1000 ref)..
34401A-34461A-34465A are all of the same "stability grade".

[djac]

This is of course correct in principle. But the rest is the practical implementation.

In the dmm noise thread only one 34470a is measured, unfortunately. With the result of 0.01ppm Vrms per range in the 10V range. The 34401a/34461a/34465a have corresponding results between 0.02 and 0.03. Since the noise of the reference is not included in this test, this means that the ADC is less noisy. It must do so in order to meet the linearity specification of the instrument, because the noise of the ADC influences and of course limits the linearity. And indeed, the linearity of the 34470a is specified as 0.6ppm and that of the others as 1.5 to 2 ppm. This is about the same factor of 3 as you find in the noise test.

On the other hand, the results of the linearyity test thread show that the 34470a does not necessarily perform better there. And the noise of the reference is included in these measurements. And the implementation of the reference on the 34470a has also been criticized (Dr. Frank, airflow of the fan over the reference, etc.).

Of course, the measurement of linearity is very demanding and therefore may be slightly inaccurate. On the other hand, even a non-optimal implementation of an LTZ1000 reference can increase the noise. So it may very well be that there is practically no big difference between 34470a and the others. But just more question marks than clarity. And then additional problems like the one of this thread unsettle all the more.

Regards Dieter

[Kleinstein]

ADC linearity and noise are not directly related. There is no real difficulty in getting low noise (about the level of the 34470/34465). However it is quite tricky to get really good linearity.  When choosing the ADCs reference current level there is even some compromising: higher current gives lower noise, but higher INL.

One major contribution to the INL is the TC and self heating of the resistor at the ADC input. This leads to a U³ contribution to the INL.  Even if the ADCs in the 34465 and 34470 are nominally identical, selecting better resistor TC can make a difference. Similar effects may apply to other parts - selected part may improve the INL or in a few cases also the noise.  Another point could be extra production-tests to verify INL for the higher specked units.

The old 34401 uses a different, older ADC type, with considerably more noise. However from the way it is build it can still have good and possibly better INL than the newer meters. The newer ADCs are mainly there for there higher speed to be used in digital RMS, not for improved INL.

Even a relatively poor designed LTZ1000 reference should get a noise level considerably lower than the LM399. Compared to the good noise level of the ADC, the LM399 can be a severe limitation when measuring a not too small voltage. However it also depends on the signal source. Unless one starts with a really low noise signal, noise from the source may still be higher than the reference noise.
The lower noise reference mainly gets important in combination with high end reference source. However the 34470 is still not made for metrology purpose as the INL specs are not that great and in theory ACAL may be limited accuracy because of the INL:

[iMo]

Anyhow, in one month it is 1 year since KS started with solving the issue.. 

[baof]

Version 3.02 still doesn't fix this problem. I think this is a hardware problem, similar to the thermoelectric potential of relay contacts, which can not be solved in software.

[Kleinstein]

Version 3.02 still doesn't fix this problem. I think this is a hardware problem, similar to the thermoelectric potential of relay contacts, which can not be solved in software.

The error during ACAL  is a strange one, so it may need quite some times to find it. My guess would be something like a thermal effect or maybe dielectric absorption, that cause trouble if readings are done with an unusual speed or sequence. So it may be really tricky to track down and maybe also difficult to fix. It looks like different meters are effected to a different extend - so some HW dependence is there.

I would not expect mechanical relays involved in the ACAL procedure (except for the high voltage ranges of cause).

[djac]

In the meantime I bought a 34470a. This one also has the "ACAL Jump", about 5µV amplitude.

Because I wanted to know if this has other consequences, I compared the 34470a with the 34465a in different aspects.

The report about it as PDF.

Regards Dieter

[djac]

In chapter "IV. Noise measuring 10V" on page 11 I noticed an error: the measurement was done with 100plc. To be on the safe side I repeated the measurement again with a shorter measuring time, so that the room temperature changed only 0.04°C during the 10 minutes. However, in this measurement the distance between the two devices turned out to be much more modest.

Regards Dieter

[splin]

An interesting report - I'm surprised it didn't attract more responses, but these are interesting/difficult times. A few points:

1) The difference in shorted input noise between the 34470A and 34465A at longer integration times is surprising. Have you compared the Allan deviance graphs to those from other posts in the DMM noise measurement thread to see if this might be specific to your samples or systemic?

You hypothesised the differences could be due to differences in autozero behaviour between the two DMMs; is this based on known differences or musings on possible causes?

2) The differences between the various shorts is also surprising - did you cover the terminals to protect them from drafts? Did you do enough repititions of each run to establish the repeatability of your tests to be sure that the results didn't reflect random drifts of the instruments rather than real differences between the difierent types of shorts?

3) On page 9, the noise measurements of an LM399 10V reference: the noise of the reference is likely 4 or 5uVpp; the .165ppm/C looks to be based on peak reference noise voltages at both the upper and lower temperatures - note the 4uV drop at 1250 seconds. Using the mean values could easily halve that TC (which should probably only be shown to one DP given the large uncertainty). Similar for the 34465A. You need a very much quieter external reference to make meaningful measurements, or at least *very* many repetitions of the tests - but you obviously know that.

4) Page 12: the 10V reference noise pp:rms ratios of the 34470A and 34465A are very different at 13.5 compared to 5.3. This surprised me; the noise from the internal and the external references are presumably Guassian, so I would expect their geometric sum to also be Guassian with a pp:rms ratio around 6. Any suggestions why the 34470A ratio is so high? In any case, the LTZ1000 noise should around 1/4 of the LM399 so should not be a material factor in the overall noise measurement.

Why are the pp measurements so similar at approx 1uV for both given that the 34465A has an LM399? I would expect the 34465A result to be almost sqrt(2) higher than the 34470A.

Do these suggest some other non-Guassian or correlated noise source in the 34470A results?

[Kleinstein]

For a noise test with an external reference one should have a really low noise reference.  So either a good LTZ1000 based one or maybe just a stable battery (e.g. in a well insulated box and mechanically protected). Ideally one would run both meters at the same time to have the same reference noise.  It makes sense to check the noise with an external voltage, as there can be other noise sources (e.g. excess noise of not so good resistors and maybe similar noise sources that change with the voltage reading)  than just the reference itself.

With LM399 there can be relatively long time scale popcorn noise. If one only looks at short windows like 100 seconds the noise reading can vary quite a bit, with some quieter and some more noisy phases.

The longer times in the Allan variance plots, where the curve goes more horizontal are likely due to thermal effects at the input / short.

The measurements of temperature effects are over a rather short time and also quite small temperature rise. It may take a little longer to really stabilize.

[djac]

Hi splin,

to 1.
If I didn't miss it, there is no measurement of the 34470a with shorted input (only with 10V input) in the "noise comparision" thread. By the way, the thread is confusing and the evaluation methods are partly different and not always clearly documented, so that a comparison is difficult.

I only tried to show a connection between the autozero mechanism and the behaviour of the 34470a, I can't say anything about the exact causes and connections.

to 2.
You are right, I did not systematically investigate the behaviour of the individual short circuit plugs, this would have gone beyond the scope and objectives of this documentation. I have only investigated which of these plugs gives a measurement that is as temperature independent as possible.
But it remains to be said that thermal voltages must be involved, because as far as I know the sockets of the 465/470 are made of copper, but the plugs used are gold plated (300nV/°C).

But the important thing is that over many measurements and all kinds of short-circuit plugs the difference between 365 and 470 remains. And this suggests that the autozero mechanism of the 470 is not as precise as the 365. It is the rapid temperature turbulence of the surrounding air which, via the very small thermoelectric voltages, causes the change from measurement to measurement in the middle range of the Allan Deviation of the 470.

to 3.
I have chosen the difference of the peaks deliberately, this results in a worst case temperature coefficient. I am aware that the absolute numbers are only correct in order of magnitude. Above all, however, it is a matter of comparing the two instruments which carried out the measurement in parallel at the same time.

to 4.
I have no better reference than the LM399 at hand. But again, it was only a matter of determining whether the better reference of the 470 is noticeable in a practical measurement. An exact measurement of the absolute value of the 470's inherent noise requires completely different measurement possibilities which are not available to me. For the rest, this point is subordinate to the objective of the documentation.

I am also aware that there are differences between the individual specimens. Therefore I would appreciate it very much if some 34470a owners could be found here, who would also do the measurements for the Allan Deviation. This could help a lot to find out the reasons for the difference to the 465. If necessary, I would be happy to provide my algorithm for calculating the Allan Deviation, so that the exact same evaluation can be done.

Regards Dieter

[Dr. Frank]

In the meantime I bought a 34470a. This one also has the "ACAL Jump", about 5µV amplitude.

Because I wanted to know if this has other consequences, I compared the 34470a with the 34465a in different aspects.

The report about it as PDF.

Regards Dieter

Hello Dieter,
thank you for your report.
It confirms my findings from 2016: https://www.eevblog.com/forum/testgear/keysight_s-new-34465a-(6-5-digit)-and-34470a-(7-5-digit)-bench-multimeters/msg889215/#msg889215
The similarities between both instruments even go further, like the same resolution of the A/D, which is fully transferred over the bus.

My conclusion at this point was, that the 34465A can be used exactly like the  34470A with 7 digit resolution, either by using the statistics function, or using scaling function, e.g. expressing the result in ppm deviation, or using readings over connectivity.

The difference in T.C. is also barely to be distinguished, as the LTZ1000A module is (like in the 3458A) probably not trimmed for minimum value, therefore could be as high as 0.3ppm/°C, which the LM399 circuit might also deliver.
Timely drift is also comparable, as the LTZ1000A circuit is operated (misused) at 95°C, like the LM399.
I received the 34470A from HighVoltage which was running continuously for nearly 1 year already, and it showed 5ppm drift in 10VDC already.
My own 34465A shows a drift of about 2 ppm after 5 years only, due to the different use case, i.e. it's only powered on when used, therefore the LM399 did not drift as much as specified (that would always imply continuous use)

Frank.

PS: This 34465A in the photograph is not yet warmed up, usually it displays exactly 10.00002V, after all these years.
Please take note of the 7 digits and of the enhanced resolution of the statistics data, simply by using scaling with M = 1 and B= 0.

[djac]

Please take note of the 7 digits and of the enhanced resolution of the statistics data, simply by using scaling with M = 1 and B= 0.

Thanks, that was new for me.

Regards Dieter

[djac]

I have revised the report once again due to the objections of splin (point 4.) In particular I increased the number of bins in the histograms to 50. Then you can see the reason why the ratio stddev/Vpp does not correspond to 6, because with the 34470a there is no normal distribution, but the drift distorts the histogram.

Regards Dieter

[HighVoltage]

Hello Dieter

Thanks for your nice reports

Very interesting to see the drift in your new instrument already now.
It will be more interesting, when the 34470A is 6 month or a year old, that is when my two instruments started to show a much larger drift from ACAL.

BTW, I have not received a final update from Keysight.

[Keysight DanielBogdanoff]

Hi all,

The R&D team and support team have been digging into this since thread popped up and have been able to pick out what's happening.

Here's the guidance from them:
"
The ACAL offset (jump) issue observed on DMM after performing ACAL can be resolved by performing full range calibration on the DC Voltage function.

The full range calibration will correct the gain ratio based on the DMM recent condition.

During ACAL, the ratios of the measurements made at a complete DMM calibration and recent measurements are used to correct the internal drift of major measurement blocks.
"

They are also digging into some other things brought up in this thread, thanks everyone for their patience!

[branadic]

Seriously? Calibrating the whole unit once ACAL was encountered is the solution provided? Wow. 

-branadic-

[Dr. Frank]

Seriously? Calibrating the whole unit once ACAL was encountered is the solution provided? Wow. 

-branadic-

Hello branadic,

no it's meant differently.
ACAL only works correctly, if a proper DCV calibration (CAL) was done before.
Once working correctly, ACAL will always work correctly, w/o another CAL.
Some units failed, for unknown reasons, maybe CAL was not done right. In these cases ACAL caused an erroneous jump up to 20ppm.
Therefore, a new CAL does cure this problem on these units.

The ACAL procedure seems to work like this:
During CAL of the 5 DCV ranges, the nominal gain factors of the ADC / REF circuit, of the 100:1 divider and of the x10, x 100 amplifier are determined very precisely .
These gain factors will drift over temperature and over time.
ACAL then precisely re-measures these actual gains and corrects to the nominal ones, by using the internal voltage reference as a stimulus.

Such 10:1 ratio re-measurements are normally not possible to do precisely enough, when using an ADC which is linear to 6 digits only.
So they probably have invented something to make this transfer 10 times more precise, at these special calibration points... maybe by additionally determining the non linearity at these fixed ADC points during the calibration procedure.

This technique is also used in a similar manner inside the 5440B calibrator, which also requires to once calibrate the nominal ratios for the different ranges.

Frank

[View[+]Finder]

I particularly dislike the fact that they placed it right in the air stream of the fan

I believe the fan exhausts, with cool air coming in the vents on either side of the case. I'm no expert on airflow in electronics cases, but logic says "take the shortest path" and there is no other "force" in the case to steer the air currents otherwise. Thus, the airstream would most likely be the result of the slight negative pressure gradient from the exhaust and the positive pressure on each of the vent holes.

[Dr. Frank]

I particularly dislike the fact that they placed it right in the air stream of the fan

I believe the fan exhausts, with cool air coming in the vents on either side of the case. I'm no expert on airflow in electronics cases, but logic says "take the shortest path" and there is no other "force" in the case to steer the air currents otherwise. Thus, the airstream would most likely be the result of the slight negative pressure gradient from the exhaust and the positive pressure on each of the vent holes.

If you have a closer look at these vent holes, the right ones give an unhindered air flow, whereas the ones on the left are mostly blocked by the inner case and the transformer.
Therefore the air goes mainly from the right holes, over the reference board area, and out to the fan, therefore nearly identical if the flow direction would be reversed. 

[HighVoltage]

Hello branadic,

no it's meant differently.
ACAL only works correctly, if a proper DCV calibration (CAL) was done before.
Once working correctly, ACAL will always work correctly, w/o another CAL.
Some units failed, for unknown reasons, maybe CAL was not done right. In these cases ACAL caused an erroneous jump up to 20ppm.
Therefore, a new CAL does cure this problem on these units.

In my experience with the two 34470A instruments, this is partially true.

Over time the ACAL drift got larger and larger.
If at that time a correct full DC calibration procedure is applied, then the ACAL function will be relative stable again.

However, after a few month time, ACAL will start drifting again.
To stabilize ACAL again, it requires another full calibration.

It seems that some instruments show a larger drift than others and some may not have this ACAL drift problem.

Full DC calibration means:

ZERO
+10 V
-10 V
+100 mV
-100 mV
+1 V

[Blue]

I'm not quite sure what equipment I should have to calibrate it fully.

Does it mean that I have to have a proper voltage reference of
ZERO
+10 V
-10 V
+100 mV
-100 mV
+1 V
?

Unfortunately I do not have proper voltage reference equipment. What to do? Send it to Keysight 4 times a year?

Any one any idea?

[View[+]Finder]

I particularly dislike the fact that they placed it right in the air stream of the fan

I believe the fan exhausts, with cool air coming in the vents on either side of the case. I'm no expert on airflow in electronics cases, but logic says "take the shortest path" and there is no other "force" in the case to steer the air currents otherwise. Thus, the airstream would most likely be the result of the slight negative pressure gradient from the exhaust and the positive pressure on each of the vent holes.

If you have a closer look at these vent holes, the right ones give an unhindered air flow, whereas the ones on the left are mostly blocked by the inner case and the transformer.
Therefore the air goes mainly from the right holes, over the reference board area, and out to the fan, therefore nearly identical if the flow direction would be reversed.

On a closer look, the holes on the left are not "blocked," there is a just redirection of the airflow to openings more to the rear of the case. There are fewer openings on the left side than on the right, true, however the airflow on the left will be warmed by the heat of the transformer as conducted through the metal case and the flow from the right will pass over warm components before reaching the reference. The key detail is that the reference is not "in the air stream of the fan," it is in the general airspace of the case some of which will be exhausted by the fan.

My experience in hot climates may be clouding my reasoning, but I recall the difference between a fan blowing directly and one exhausting air with replacement through open windows. If this bears on the success of my reference transplant, it would be a fun project to put some sensors in the case and run some tests. Let me know . . .

[ScoobyDoo]

These (high) drift figures for 34470A were recently reported by maxwell3e10 in post:

https://www.eevblog.com/forum/metrology/usa-cal-club-round-2/msg3552046/#msg3552046

Best regards
ScoobyDoo

[Kleinstein]

The ACAL change shown in the graphics is different from the 34470 specific ACAL problem, that is more transient: so after ACAL there is additional settling for some time (~ 1 Hour ?). The reason behind this still seem a bit unclear and the fix from KS looks a bit odd, likely not really solving the root cause, more like a work around. Different meters seem to be effected to a different degree.

The graph shows how much change from calling ACAL after a very long time of not doing so. This is more like long time drift of resistors that are not made to be long time stable. In addition the times are different and the degree of what is corrected by ACAL can also vary. So the meters are only partionally comparable.

[HighVoltage]

For me, the comparison reference is the Keithley DMM7510. It is always spot on stable, even if the environment temperature is changing and it does NOT need the ACAL to be pressed at all, although it has an ACAL function.

The 34470A is most stable, if the ACAL is not pressed at all.
However, it will drift over time.

Where the 34470A is really shining bright is the low current range.
And here it is better than the DMM7510.
So, at the end of the day we need both instruments.
I still like the 34470A but one needs to be aware of its limitations.

[Kleinstein]

The 34470 may very well also be quite good with AC - though probably not much difference to the 34465 in this respect.

The Keithley meter seem to do some real time measurment and correction of the ADC gain, so kind of doing a partial ACAL in the background all the time. However chances are this is relatged to the odd extra noise one gets with most Keithley meters at some 100 PLC or 10 s time scale. So the low drift comes at a price - though this is likely more like a software problem, not an absolute technical requirement. So it looks like the KS34470 is not alone with a slow to fix software problems effecting the performance.

The real time correction likely does not include the amplifier gain, so one may still want an occasional ACAL for the 100 mV and 1 V ranges with the DMM7510. The 10 V range likely would not need it at all.

With the partial "fix" from KS, not sure if they will find the true reason behind the settling after ACAL and fix it.

[HighVoltage]

The ACAL function of the Keithley DMM7510 is hiding in the menus.
But each time I have used it, it had almost no effect at all on both of my units.
So, from my experience, one does not need it at all.

Actually, you are right, the 34470A is very good in the AC ranges but I usually have no need for that and therefore I have no deep experience with it.

Most impressive is the low range DC current. Here I am measuring a 1n source with the 34470A. Even over longer time, this is very stable.

[HighVoltage]

Keysight released a new Firmware version 3.03 on 2021-08-25

Today I installed this version 3.03 on my 34470A but unfortunately the ACAL problem was not addressed.
It seems like other important issues have been addressed.