Hi,
I got my brand new 34465A DMM last week, it had been calibrated by Keysight (initial/ production calibration) June 1st.
Out of the box, it had a disturbingly high calibration count of 222.
Another owner of a new 34470A found 76 cal counts, which seems also too high, maybe up to 20 or 40 seems to be reasonable, as the count seems to increase by 1 per calibration group only. (calibraton of 100V and 500V DCV increases by 1 only)
I assume, Keysight currently has big problems to finally calibrate these instruments in their production run.
I found out, that the DCV ranges 100mV, 1V, and +/- 10V were spot on, after using ACAL, always < 1ppm accurate compared to my Fluke 5442 and HP3458A.
The 100V and 1000V ranges were both off by +14 and +17ppm (this can be checked precisely relative to the lower ranges by the 3458A, at least).
I could not understand why the lower ranges were in accordance to the test report (also indicating <1ppm deviation) , whereas 100V and 1000V should have drifted 14 and 17ppm in these 3 weeks, as the instrument was no used in that period of time.
So I recalibrated 100V and 1kV, and it turned out, that the following readings (verification) were indeed spot on (< 0.3ppm deviation) , as expected, even after cycling power.
As soon as I engaged ACAL, the +14ppm and +17ppm deviation was introduced again.
This is very probably a SW bug, which does not implement T.C. correction for the 100:1 divider in the correct manner.
Maybe the routine misses the 100:1 attenuation..
If I recalibrate the 1kV range only (500V applied), this will already correct 1kV and 100V range, i.e. it's not necessary to recalibrate 100V.
There is a second (software) bug:
In 100V and 1kV range, the reading is correct, when the input impedance is set to AUTO (i.e. > 10GOhm).
If I switch to 10MOhm, the reading is +3ppm higher in both ranges, although there is no hardware switching involved, as the HV ranges are already in this 10MOhm mode!
If I calibrate with 10MOhm setting, the verification also fails by +3ppm, switching to AUTO removes this error.
Both errors sum up to +17 and +20ppm, which already is the 90 day drift.
During Keysights calibration process, as ACAL probably is engaged before verification, that will probably cause a calibration error, explaining the high 222 counts of my instrument.
Secondly, this instrument will fail the 90 days or 1 year specification, if it already starts at 20ppm and the normal drifts will add to that.
So I need support from Keysight engineering, to get a confirmation of that bug, and a fix in the firmware, or in hardware, if it turns out to be HW related (less probable to my understanding).
I could also supply serial number, test report and further details of my own measurements.
Regards
Dr. Frank
I got my brand new 34465A DMM last week, it had been calibrated by Keysight (initial/ production calibration) June 1st.
Out of the box, it had a disturbingly high calibration count of 222.
Another owner of a new 34470A found 76 cal counts, which seems also too high, maybe up to 20 or 40 seems to be reasonable, as the count seems to increase by 1 per calibration group only. (calibraton of 100V and 500V DCV increases by 1 only)
I assume, Keysight currently has big problems to finally calibrate these instruments in their production run.
I found out, that the DCV ranges 100mV, 1V, and +/- 10V were spot on, after using ACAL, always < 1ppm accurate compared to my Fluke 5442 and HP3458A.
The 100V and 1000V ranges were both off by +14 and +17ppm (this can be checked precisely relative to the lower ranges by the 3458A, at least).
I could not understand why the lower ranges were in accordance to the test report (also indicating <1ppm deviation) , whereas 100V and 1000V should have drifted 14 and 17ppm in these 3 weeks, as the instrument was no used in that period of time.
So I recalibrated 100V and 1kV, and it turned out, that the following readings (verification) were indeed spot on (< 0.3ppm deviation) , as expected, even after cycling power.
As soon as I engaged ACAL, the +14ppm and +17ppm deviation was introduced again.
This is very probably a SW bug, which does not implement T.C. correction for the 100:1 divider in the correct manner.
Maybe the routine misses the 100:1 attenuation..
If I recalibrate the 1kV range only (500V applied), this will already correct 1kV and 100V range, i.e. it's not necessary to recalibrate 100V.
There is a second (software) bug:
In 100V and 1kV range, the reading is correct, when the input impedance is set to AUTO (i.e. > 10GOhm).
If I switch to 10MOhm, the reading is +3ppm higher in both ranges, although there is no hardware switching involved, as the HV ranges are already in this 10MOhm mode!
If I calibrate with 10MOhm setting, the verification also fails by +3ppm, switching to AUTO removes this error.
Both errors sum up to +17 and +20ppm, which already is the 90 day drift.
During Keysights calibration process, as ACAL probably is engaged before verification, that will probably cause a calibration error, explaining the high 222 counts of my instrument.
Secondly, this instrument will fail the 90 days or 1 year specification, if it already starts at 20ppm and the normal drifts will add to that.
So I need support from Keysight engineering, to get a confirmation of that bug, and a fix in the firmware, or in hardware, if it turns out to be HW related (less probable to my understanding).
I could also supply serial number, test report and further details of my own measurements.
Regards
Dr. Frank
yes, Keysight very quickly reacted.
They've sent me a new instrument, and the "old" one is now on its way to Malaysia, to be analysed.
They already confirmed the +3ppm shift on other instruments, but try to assign the other fault to thermal effects (by 1000V burden).
I think, I already have busted their argument, so they should investigate further.
My new 34465A has 76 cal counts only, and this is the default count if being fresh from the factory.
Of course it has this +3ppm error also, and the 100V range also shows + 10..+ 14ppm shift relative to its 3 weeks old calibration / validation, compared to the 100mV, 1 and 10V ranges, which were again spot-on, i.e. deviation < 1ppm (+ thermal offset in the 100mV range).
I did not want to confirm this calibration fault again, by re-calibration of the 100V and 1kV ranges, up to now.
So I really have more and more proof, that also the calibration shift by ACAL is a real firmware issue.
In the end, I am very satisfied with the support and effort by Keysight!
Frank
thank you very much for your feedback.
My new 34465A has a cal count of 82, obvoiusly fresh from the factory.
I bought this DMM because of it's accuracy (in relation to the price...) an Keysight's reputation.
So, i hope they will fix this bug asap (and no recalibration is necessary...).
Thanks again.
Engage statistics, so 7 digits will be displayed as average.
When you toggle the impedance between 10M and AUTO, the value should change by that amount (always clear statistics)
If your instrument is brand new, meaning that the calibration is only a few weeks ago, you may also check for the other possible error:
Apply a stable 10V in the 10V range and measure its precise value, to 6 or 7 digits (with statistics).
Now switch to 100V, impedance AUTO, and measure this same value to 6 digits resolution, using statistics/average.
As the A/D converter is linear to typical < 1ppm, a possible deviation of 10ppm should be easily detectable.
10V and 100V should have been validated to <= 1ppm deviation, as documented in your calibration test report.
That means, the 10:1 ratio (10V range vs. 100V range) is also precise to about 1ppm, initially.
If you find a bigger deviation, that's a hint, (no certain proof!) that your instrument also got this error.
Frank
from Frank's post >>There is a second (software) bug: In 100V and 1kV range, the reading is correct, when the input impedance is set to AUTO (i.e. > 10GOhm). >>If I switch to 10MOhm, the reading is +3ppm higher in both ranges, >>although there is no hardware switching involved, as the HV ranges are already in this 10MOhm mode!
Just tried this second bug on my early S/N 34461A (no Auto Cal). Using a Datel DVC 8500 set to about 19.9V on the 100 V scale, when I switch from AUTO to 10 M, (using stats, about 50 measurements) I see a consistent increase of about 60 uV (+6 LSD of the Average), or +3 ppm.
Easiest to do by using 50 pts acquisition, DCV 100 V scale, stats on, single manual trigger:
19.899 92 - 10M
19.899 85 - Auto
19.899 91 - 10M
19.899 84 - Auto
19.899 91 - 10M
19.899 84 - Auto
The 34461A is "drifty" at about 1ppm / c with room temperature, so the 1 ppm shift is expected with the lab windows open and a breeze in the room.
Same experiment 34410A (the "10A") between 10 M and Hi Z, no change or 1 LSB, but was only warmed up for a short time.
Edited by: tayezisk on Aug 23, 2015 12:04 PM
No other DMM shows such an effect, (34401A, 3458A) of course, because there is absolutely no logical circuitry explanation for such.. Also see the schematics of the 34410/11A, which should be very similar..
The drift parameters all are not relevant, as you can more quickly switch between AUTO and 10M setting, and so quickly read with and w/o this 3ppm offset, as you nicely demonstrated.
Frank
here's the end of this story:
Scott Stever, previous PM of the 3458A, and now also involved in the 34465A/470A development, explained to me in a phone conference the characteristics of the ACAL function.
It's a pity, but he could not go into details, how this is realized in hardware.
This function is able to make small corrections to all the gain constants of volt and ohm modes, obviously by re-measuring the ratio factors, which were initially measured in the fab.
This works very well in the Ohm and low DCV ranges, after 3 months; I observe that all are still within 1ppm of the initial values.
For the the HV range, as there is this big 100:1 attenuation factor and a higher noise level, the function is limited to about 5..10ppm repeatability.
This is the approximate amount of deviation, the ACAL function will introduce basically for the HV DCV ranges.
This is well inside the 90d specification limits, and also included in the 24h specification.
The overall specification calculus, including the T.C., is greatly improved by the ACAL function, but not to the same level like the "real" ACAL function of the 3458A.
So the expectations should not be too high, when users of the 34465A/470A already have experience with this 3458A metrology grade instrument.
Also, ACAL is not needed on a daily basis, and should only be used, when the room temperature changes by more than about 5°C.
As Steve Scott explained to me, the 34465A and 34470A are superior in total performance over the previous 6 1/2 digit DMMs, like the 34401A and 34410/411A.
I agree to that, especially as the HMI is much better, and the most used function finally are available in one keypress.
In retrospective, the people at Keysight have done a great job on my issues.
They really have taken care of my concerns, even when I insisted on explanations about the ACAL peculiarity.
They gave a great support, including a new instrument!
A new FW for the +3ppm failure will be released in the next days / weeks, it's currently undergoing their tests.
Frank