34401A resistance ranges stability

[splin]

There is plenty of evidence that old 34401As can have very stable 10V ranges due to the stability of aged LM399s. But what about the resistance measuring ranges? Does anyone have any data or links to 34401A calibration results from which it is possible to get an idea of typical annual drift rates? 

[dacman]

Older 34401As with a square power button need to be adjusted less often than newer 34401As with a round power button, but it is usually due to ACV.  The older units do need to be adjusted sometimes, but nothing like the newer ones.

[cellularmitosis]

Here's the ohms current source (attached).

Two of the important resistors are U102 and R202.

U102 is a bunch of resistors inside of a ceramic chip.  I would expect it ages very slowly.

R202 (40k) is a Vishay S102C metal foil resistor.  This is in an epoxy package, which is affected by seasonal humidity changes.

I'm not sure what R201 (400k) is.

[Kleinstein]

R201 is not that critical: it is used only for the highest range, where stability is more effected by semiconductor leakage than resistor stability anyway.

[bitseeker]

Older 34401As with a square power button need to be adjusted less often than newer 34401As with a round power button, but it is usually due to ACV.

Odd. Why's that?

[splin]

I was hoping for some typical numbers, perhaps from someone who calibrates them or had some calibrated regularly?

My guess is that typical drift for the 1K range will be around 10 to 15ppm/year (compared to the datasheet 100ppm maximum) but I have no evidence.

[try]

Hi splin,

There is plenty of evidence that old 34401As can have very stable 10V ranges due to the stability of aged LM399s.

what is "very stable" in quantitative terms?

Regards
try

[splin]

Hi splin,

There is plenty of evidence that old 34401As can have very stable 10V ranges due to the stability of aged LM399s.

what is "very stable" in quantitative terms?

Regards
try

Sorry I can't give you any proper statistical data, but there is anecdotal evidence. See reply #9 for example:

https://www.eevblog.com/forum/testgear/34401a-long-term-accuracy-question/

After 13 years it's still within the 12 month spec on all ranges, and 5x better on most DC voltage, current and resistance ranges.

And from http://www.candlepowerforums.com/vb/showthread.php?418230-Checking-my-DMM-DC-Voltage-Accuracy&styleid=32

What I was extremely lucky on was the Agilent 34401a. Its spec is 15 PPM for one day and 35 PPM for one year. My unit is 15 years old and appears to have the original factory calibration. It passes all self tests and has no reported errors. It had been certified several time but apparently never re-calibrated. I have never observed any error greater than 10 PPM even at turn on. With 15 minutes of warmup the error is usually less than 4 PPM.

FWIW, my own 34401A, which hsn't been calibrated for at least 8 years exactly matched, to the uV, that of a cheap Chinese AD584 calibrator which supposedly was measured by the vendor with a calibrated 34401A. I thought that was remarkable but it doesn't prove anything of course.

Also two posters here have LM399 based references which they state have annual drift rates of less than 1ppm. Another post here stated that  the LM399 has essentially zero drift when unpowered but I don't know of any tests to confim or deny this. This leaves the 7 to 10V scaling resistor network as the likely main source of reference drift in the 34401A.

The best data would come from someone who calibrates these meters for a living but I don't know if we have any posters here who does that and is prepared to provide it.

 

[cellularmitosis]

I believe it was Dobkin who said the 399 doesn’t drift while powered off. If you look at the chart in the LTZ1000 spreadbury paper, it appears they do not drift appreciably either when powered off.

[borghese]

Yes, was Bob. See my replay #720 and #722 on " LM399 based 10 volt reference" topic.

[Kleinstein]

For the resistance measurements the voltage reference has essentially no effect.  It is about the drift of meter internal resistors. For the higher ranges (e.g. MOhms) also changes in leakage currents could also be a factor. The relevant resistors are R202, the ratio of resistors in U102 and also the gain stage for the smaller resistors. I don't have number, and the results from just a single or a few meters would not be really relevant. It  would need data on quite a few meters to really get a meaningful statistics. So some of the cal labs might have those data.

[AG7CK]

I should not really post here, because a Fluke 8505A is not a HP34401A  .

However - after Kleinstein's excellent post over leaves LM399 related posts somewhat off target, I feel compelled to give another example of what he is talking about.

In the 8505A (Ohms Converter module) resistance measurement is totally decoupled from reference value and long time drift. Only short time noise matters, and that can to some extent be averaged out.

This figure from the manual shows that resistance is calculated as [Voltage over Rx / Voltage over Rref] for the same current. The result is independent of the reference (as long as the ref and ADC behaves good the few milliseconds or so it takes to do two measurements).

[try]

Hi Splin,

Hi splin,

There is plenty of evidence that old 34401As can have very stable 10V ranges due to the stability of aged LM399s.

what is "very stable" in quantitative terms?

Regards
try

Sorry I can't give you any proper statistical data, but there is anecdotal evidence. See reply #9 for example:

https://www.eevblog.com/forum/testgear/34401a-long-term-accuracy-question/

After 13 years it's still within the 12 month spec on all ranges, and 5x better on most DC voltage, current and resistance ranges.

I am a simple person. I expect ppm terms! 
Being within 12 month specs is not that difficult and "5x better on most...", well, "most" does break my calculus.

And from http://www.candlepowerforums.com/vb/showthread.php?418230-Checking-my-DMM-DC-Voltage-Accuracy&styleid=32

What I was extremely lucky on was the Agilent 34401a. Its spec is 15 PPM for one day and 35 PPM for one year. My unit is 15 years old and appears to have the original factory calibration. It passes all self tests and has no reported errors. It had been certified several time but apparently never re-calibrated. I have never observed any error greater than 10 PPM even at turn on. With 15 minutes of warmup the error is usually less than 4 PPM.

Against his Asia-Volt?
In the text you referenced they mention having two references with the same calibration figures. That would make me suspicious.

FWIW, my own 34401A, which hsn't been calibrated for at least 8 years exactly matched, to the uV, that of a cheap Chinese AD584 calibrator which supposedly was measured by the vendor with a calibrated 34401A. I thought that was remarkable but it doesn't prove anything of course.

Also two posters here have LM399 based references which they state have annual drift rates of less than 1ppm.

You are comparing the 10V range of a 34401A against a LM399 based reference (7V or 10V or another output figure?).
This appears like comparing apples and pears to me.

Another post here stated that  the LM399 has essentially zero drift when unpowered but I don't know of any tests to confim or deny this. This leaves the 7 to 10V scaling resistor network as the likely main source of reference drift in the 34401A.

What is "essentially zero"? That could have taken from a Vishay datasheet. "0.1" can be considered as "essentially zero" as well.

The best data would come from someone who calibrates these meters for a living but I don't know if we have any posters here who does that and is prepared to provide it.

Anyway, you seem to be the right candidate to join the maker faire 2018 here in Hannover, Germany on the 15. and 16.9.2018. Going there is my top priority.
It is very easy to get there, something fellow eevblog member Tom from the United Kingdom who came here last year could confirm you.

You can calibrate your device yourself at the PTB booth with the only disadvantage being faced with higer uncertainties due to the uncontrolled environment.

For reports in German language describing the last two events check the following link out and look for "r2d3".
As for r2d3's LM399-based 10V reference you can see on the pictures that it moved 3 ppm from 2016 to 2017 in 16 months. In 2017 it was 10V spot on but that does not say alot given the uncertainty of the 3458A and the conditions involved on the fair in 2017.
During this 16 months r2d3's 34401A moved a lot more (any quantitative terms omitted on purpose to fit the forum style  ) than his LM399-based reference.
Both instruments ran 365/24 between the two events. You might consider power-on time as an influencing factor as well.

https://www.mikrocontroller.net/topic/398160

Have fun!

Regards
try

[Jon.C]

a question about the 34401A

how many counts does it have?

I have read about display update speed and other features ... etc ..

https://www.edn.com/electronics-news/4389451/What-s-a-half-digit-anyway-



"Consider the Hewlett-Packard 34401A as an example. HP says the meter has a 6½-digit display. By applying the fractional rule, you would predict that the meter has a maximum display of 1,999,999 counts. The 34401A's specs, however, show that the meter's full count is 1,200,000. If the HP 34401A really a 6½-digit meter despite its counts? Yes; each of the full digits can display the full range of 0 to 9, and the ½-digit displays 0 or 1. For example, a 99.9999 reading on the HP 34401A is possible, but a 199.9999 reading is not. Therefore, you can't assume that a meter with a ½ digit will have full-scale readings of 199.9V, 1.999V, or 19.99V."

[cellularmitosis]

That's correct, the 34401A and similar Keithley models (2000, 2015) have "20% overrange", i.e. 1.2 million counts.

https://github.com/USACalClub/gear/blob/master/dmms/README.md

[Kleinstein]

Those meters with 1.2 million counts are more like a 6 digit class, especially as the count beyond the 1 million counts are only over-range.  Formally the display is 6 1/2 digits, though 6.1 digits would be more honest. Anyway with meters of this class one generally looks at the specs before buying. So marketing phases are not that important here.  Normally with a 3.5 digit meter one can assume a +-1999 range, similar with 4.5 digits.

Having ranges to something like +- 12 V or similar is a kind of a natural limit found with many meters, as many OPs and CMOS switches are made to work with a +-15 V supply and something like +-12-13 V is the maximum amplitude to be easily handled with this supply.  As a funny side node the 34401 ADC is a little odd that is more like having a +-3 V range and uses a kind of divider (resistors to input and GND) to get the +-12 V range.

[Andreas]

As a funny side node the 34401 ADC is a little odd  that is more like having a +-3 V range

no.
(  ) you know how a integrator works (otherwise go back to school until you can tick the box).

Cannot contribute with values but:
our calibration department guy stated once that the resistance range of a K2000 is typically (much) more drifty than that of a 34401A.

from old Keithley homepage I have the info that usually the devices are not adjusted if they are within 70% of the 1 year spec.
my conclusion: The drift after the first year is less than 30% of the 1 year spec.

with best regards

Andreas

[Micke]

Regarding counts and number of digits, interesting read in Keysight app note on 34410A
https://literature.cdn.keysight.com/litweb/pdf/5989-4879EN.pdf?id=819561
Surprisingly the A/D can measure to 14V, but artificially limited to 12V, to reduce nonlinearity in end span I guess?

From app. note:
Companies do not usually pub-
lish the internal span of their
A/D converters, so it may not
be possible to derive this num-
ber when you compare specs
from one company to another.
In the case of the 34410A, the
internal range is ±14 V on the
10 V range, even though the
usable range is artificially lim-
ited to ±12 V. So, plugging in,

ENOD = log10(14/.0000082 ) =
6.2 digits (68.3% probability)

or

ENOD = log10(28/.0000455 ) =
5.8 digits (99.7% probability)

Since the spec is based on rms
noise, not peak-to-peak noise,
we would call this a 6 1/2-digit
DMM on the 10 V range at 1000
readings/s. The same method
can be used to calculate the
ENOD for other ranges and
speeds.

[splin]

As a funny side node the 34401 ADC is a little odd  that is more like having a +-3 V range

no.
(  ) you know how a integrator works (otherwise go back to school until you can tick the box).

If you really want to split hairs, according to my calculations the 34401A integrator output swings by a maximum of 4.04V p-p, centered at approx 2.8V. The input to the residual ADC is clamped to 0V min and 5V maximum.

The Art of electronics 3rd edition describes the Multislope III ADC as used in the 34401A and 34420A on page 919 but they seem to have got it wrong. Footnote 54 states:

The 'ADC's conversion range matches that of the integrator, but both are ~8x larger than the integrator's ramp over a clock tick when the input signal is +/-full scale. That is why the ADC effectively loses 3 bits of resolution when digitizing the residue (Vf - Vi)

The maximum ramp is 2.75V in one clock tick so the factor is actually 5/2.75 = 1.8x. However they also believe the residue ADC is 12 bits whereas as far as I can see it is actually the 10 bit convertor integrated into the 80C196KB. Thus the answer comes out the same, approx 9 bits of resolution for the residual ADC, but for the wrong reasons.

I also can't understand why HP increased the reference voltage from 7 to 10V rather than using different values for the current steering resistors in the ADC input - it seems to be entirely unnecessary and a poor choice from a stability point of view. I don't believe the +/-10V references are used anywhere else in the instrument.

Cannot contribute with values but:
our calibration department guy stated once that the resistance range of a K2000 is typically (much) more drifty than that of a 34401A.

from old Keithley homepage I have the info that usually the devices are not adjusted if they are within 70% of the 1 year spec.
my conclusion: The drift after the first year is less than 30% of the 1 year spec.

with best regards

Andreas

Thanks, that's interesting as the only data point posted so far.