Fluke 5440A calibration

[Rax]

I've done a couple of power cycling experiments. As expected, the 5440A has about four hours warmup before it's fully stabilized - exactly as much as Dr Frank indicated - while the Prema 6048 seems to be fully up in maybe about an hour. At the end of that, they're in the exact same agreement as before. The 731B I didn't power cycle, but the agreement stays exactly as tight with it also.

Other than energy savings, I am not convinced I should continuously power cycle these every day. I tend to use them daily, and to me the frequent power cycle - with the oven cooling down and then warming up again - seems more radical and introducing more wear than keeping them on. I do think they're in the part of their career where there's very little aging to still occur.

On our different takes above, alm - to me, it's pretty obvious we're both correct. We just adopt points of view from two different paradigms, and have not truly shifted our universe/domain of discourse to really meet. You, as a metrology guy (I assume), adopt a mindset where statistics dominates all the way to an ontological relevance. I, as a physicist, set limits to what statistics can say will or will not happen.

I appreciate your points and have carefully read all the resources and articles you've presented here.

[Rax]

On our different takes above, alm - to me, it's pretty obvious we're both correct. We just adopt points of view from two different paradigms, and have not truly shifted our universe/domain of discourse to really meet.

Let me flesh that out a little more. Typically what this paradigmatic shift means in physics is, if you apply a limit condition to a certain paradigm, it needs to blend seamlessly into the other paradigm for both to be correct (see Special Relativity vs. Newton's Mechanics).

In our case, with your "highway speed" point, alm, you assume all of these decades-old instruments still have relatively large (or: "significant") drifts. And maybe they do - I don't have many years experience in this field - but I have not seen them drift in months' long, daily observations accruing hundreds of hours (as I said). I may still be wrong (maybe my methodology is faulty), but if they still significantly drift, I should either observe it, or they all drift exactly the same (which I think is impossible, as they don't share the same references, design is completely different, even their age is likely different, etc.).

So if you take your assumption of drifting and instead assume it's insignificant, my point on accuracy and reliability of my bench volt becomes valid.

To bring a metrology point in this conversation, let's look at what Solartron-Schumberger says relative to their 7081 8.5 digit meter (see attachment). I'd love to hear others' experience with drifting references (particularly decades-old), but SS states past nine years the drift is "insignificant." This is on a chart including sub-ppm quantifications, and for an instrument qualified to measure sub-ppm quantities.

[Rax]

SS states past nine years the drift is "insignificant."

I think the 5440A was made in 1981, so nine years past that is in 1990. We're 30 long years further down the line past that.

I should also say, I've just probably just contemplated a quick, 3ppm drift from a brand new (probably aged) Fluke reference in a recent production Fluke meter.

[MegaVolt]

I should also say, I've just probably just contemplated a quick, 3ppm drift from a brand new (probably aged) Fluke reference in a recent production Fluke meter.

When comparing old and new devices you should take into account the change in the volt standard in 1990 to 9.3ppm

[Rax]

When comparing old and new devices you should take into account the change in the volt standard in 1990 to 9.3ppm

Great point! I never quite got to a point where I feel I incorporated that into all this.

Does that mean all calibrations prior to that change's date are off by today's standards?

[MegaVolt]

Does that mean all calibrations prior to that change's date are off by today's standards?

You must take into account the correction if you take old data. This is a simple arithmetic correction.

[Rax]

I tried to shake my 5440A and the Prema 6048 out of agreement (curiosity?...), and I must confess I haven't had much success.

I kept them off for a day or two now and then (different times!), not run INT CAL, or zeroing, and then did, vary temperatures at the bench a bit wilder than before (plus airflow). I was able to run a slightly over 1ppm disagreement by running the Prema on its back (seems pretty impactful, BTW), and varying the temperature gradients in the garage pretty wild (door open to allows drafts to shower the instruments, etc.). The latter is also pretty impactful.

But I guess they just designed both the 6048 and the 5440A too well... The most I could shake them off balance is by a bit over 1ppm. Of course, this says nothing about long term this and that (or absolute accuracy), but just to assess short term variations on ovenized instruments. Variations turn out be difficult to quantify (being as small as they are).

Once I ran both instruments a bit more civilized - constant/stable, though admittedly high - about 27C - temperatures, though their evenness helps zeroing on the Prema and the INT CAL on the Fluke deal with it - they settled back to under .2ppm agreement.

[Rax]

About those 27C... In case that was a point of contention, I am at 22.8C right now and see the exact same agreement between the Prema 6048 and the Fluke 5440A. On the dot 10.0000000V, +-.05ppm all night tonight.

Hopefully, this is not construed as obscene.

[Dr. Frank]

About those 27C... In case that was a point of contention, I am at 22.8C right now and see the exact same agreement between the Prema 6048 and the Fluke 5440A. On the dot 10.0000000V, +-.05ppm all night tonight.

Hopefully, this is not construed as obscene.

Have you already checked their agreement on the other ranges?
Frank

[Rax]

Have you already checked their agreement on the other ranges?
Frank

Yes, but not as systematically (though I must have some notes somewhere). But for instance, this morning - they've been running overnight - at 25.6C I found the P6048 indicating about .2ppm high at 10V (fluctuating between .15ppm and .23ppm from integration to integration).

And then, if I apply:

• 100V: 99.998724V
• 1000V: 999.98352V

I am not positive I gave them enough time - I think there's a bit of settling time needed for higher voltage measurements - as I'm right before a flight to a conference across the States, but that's probably close enough to what I've seen those ranges typically doing.

[Mickle T.]

Is there any chance we'll see short-term stability charts for the P6048+F5440A instead of a bunch of words about ppm's? 

[Rax]

Is there any chance we'll see short-term stability charts for the P6048+F5440A instead of a bunch of words about ppm's? 

I'd love to be able to do that - and I've had some conversations on the P6048 thread on that topic - but I'm not currently able to read my P6048 over GPIB to automate collection of data and plot some charts.

[alm]

You can make a chart the old school way by manually registering measurements, ideally something like 16 values per point so you can get a decent estimate of standard deviation, and then plotting the results.

[Rax]

You can make a chart the old school way by manually registering measurements, ideally something like 16 values per point so you can get a decent estimate of standard deviation, and then plotting the results.

Here's a shot at that. Temperature is, unfortunately, 27.4C (I was at a conference in MN over an extended weekend of sorts, and was so happy to come back to 85F in SoCal from the high 20s!....).

Both the 5440A and the P6048 have been off for about a week. I had my better half turn both on this morning (while I flew all day, so I couldn't handle that), to make sure they both plateau thermically ("warmup," of course). Upon arriving home in the afternoon, I only zeroed the P6048, nothing else (though it'd be interesting to see the effect of INT CAL on the 5440A). There's always a couple of readings initially that are a bit off, and so below I just started noting down at the point that the readings stabilized. Each one of the readings below is a consecutive 20s integration by the P6048.

• 9.9999999V
• 10.0000002V
• 10.0000006V
• 10.0000002V
• 10.0000003V
• 10.0000001V
• 10.0000002V
• 10.0000005V
• 10.0000008V
• 10.0000007V
• 10.0000005V
• 10.0000005V
• 10.0000004V
• 10.0000004V
• 10.0000003V
• 10.0000004V
• 10.0000004V
• 10.0000004V
• 10.0000004V
• 10.0000008V
• 10.0000012V
• 10.0000012V
• 10.0000016V
• 10.0000018V
• 10.0000018V
• 10.0000015V
• 10.0000016V
• 10.0000017V
• 10.0000016V
• 10.0000015V
• 10.0000011V
• 10.0000012V

[here I had to deal with some Halloween guests, so I missed a few integrations]

• 10.0000002V
• 10.0000005V
• 10.0000003V
• 10.0000000V
• 10.0000001V
• 10.0000001V
• 10.0000003V

.....

[Rax]

...I kept on going, but didn't see anything more than .13ppm of 10V on any integration I observed. It may wonder on the high end of a .01ppm variation from 10V (meaning, not on the 9.9999999V side), and then creep up to wondering around a .1ppm variation of the same. In that sense, that's a "high" variation. It's typically more stable than this on a given evening, but I blame the time they both were off with its moisture accrual, wet caps getting a bit wacky, etc., and also just general environmental conditions fluctuations in an unconditioned space.

[Rax]

I went ahead and also did the INT CAL and it seems, if anything, the agreement between the 5440A and the P6048is may be a bit tighter. For one, I don't seem to see those .1ppm excursions. I see it fluctuating in the ".01ppm range" (last count on screen) from 10V.

[Rax]

Is there any chance we'll see short-term stability charts for the P6048+F5440A instead of a bunch of words about ppm's? 

I was asked for a chart, so here's a quick one based upon the same readings. Note the first steeper slope is due to the units still stabilizing/establishing their first reading(s), while the second steeper slope on the downward trend this time is the jump from a "12" to a "02" which is much less steep (it occurred over several successive integrations, as I was away from the bench and dealing with guests).

[Dr. Frank]

Hello Rax,
what's the scaling and the unit of the X-axis?
Is it possible for you to perform a stability measurement over 12 or 24 hours?
You might monitor the environmental temperature as well.

Frank.

[Rax]

Hello Rax,
what's the scaling and the unit of the X-axis?
Frank.

Frank,
Every single point on the X axis is a successive reading at the end of a 20 seconds integration. On the Prema 6048, that's the shortest integration that brings all digits on the screen (8.5). So, essentially, X axis is time, discrete/quantified to 20 second units, if you will.

[Mickle T.]

It makes sense to increase the integration time to 80 seconds.

[Rax]

Hello Rax,
Is it possible for you to perform a stability measurement over 12 or 24 hours?
You might monitor the environmental temperature as well.
Frank.

On both of those accounts - not really. All this is manual harvesting of data, as I explained (I haven't yet been able to work out GPIB or other ways to automate my data harvesting). So to do 12 or 24 hrs continuously would require me to take vacation from my actual work, including time past those to 24 hrs to catch up with sleep!...

Same applies to temperature, I can write down what my bench thermometer tells me every time I collect a reading, but it doesn't sort out the issue of "sweatshopping myself...."

That said, I've found the instruments indicating about 10.0000020V this morning (at the end of a warm October night in SoCal) and I have no reason to think the overnight variations exceeded that (in my experience, largely speaking, when talking about an unconditioned garage, 5AM is usually the most divergent environmental condition of the night). And as environmental conditions in the room became more regular (with my presence in the room, etc.), the readings have gone closer to 10.0000000V than that. I've seen small variations of the last count as I had my first meeting of the morning.

[Rax]

It makes sense to increase the integration time to 80 seconds.

I haven't personally seen the readings change much, if all all, by doing that, and it just makes collections of data painful and impractical. I'd almost rather stick with 20s and collect every fourth one... But is there another specific reason you think I should switch to 80 seconds? Tighter integration?

[Mickle T.]

I suggest increasing the integration time, since your multimeter is quite noisy (not least due to the ICL7652 in the input amplifier). Based on the 10-minute chart you presented, it is generally not clear how you can talk about some *.01 ppm, when even 10-minute instability is an order of magnitude higher.

[Rax]

I suggest increasing the integration time, since your multimeter is quite noisy (not least due to the ICL7652 in the input amplifier).

That makes sense, thank you.

Based on the 10-minute chart you presented, it is generally not clear how you can talk about some *.01 ppm, when even 10-minute instability is an order of magnitude higher.

I explain that here, hopefully better:

I went ahead and also did the INT CAL and it seems, if anything, the agreement between the 5440A and the P6048is may be a bit tighter. For one, I don't seem to see those .1ppm excursions. I see it fluctuating in the ".01ppm range" (last count on screen) from 10V.

Meaning, typically, and for relatively short stretches of time, the fluctuations are restricted to the "last count" on the 8.5 digit screen of the P6048. More importantly, before any critical work, I feel I can get these to agree very, very closely by running a fresh zero on the P6048 and an INT Cal on the F5440A. I wager that may also mean "more accurately," but that's a debatable point.

But stepping back for a larger perspective over things, I think this is a fair statement for what I'm seeing and is confirmed days later by the chart I just posted:

Once I ran both instruments a bit more civilized - constant/stable, though admittedly high - about 27C - temperatures, though their evenness helps zeroing on the Prema and the INT CAL on the Fluke deal with it - they settled back to under .2ppm agreement.

[alm]

On both of those accounts - not really. All this is manual harvesting of data, as I explained (I haven't yet been able to work out GPIB or other ways to automate my data harvesting). So to do 12 or 24 hrs continuously would require me to take vacation from my actual work, including time past those to 24 hrs to catch up with sleep!...

Okay, if you're not set up for short term drift measurements, how about long term drift?

• 9.9999999V
• ...

Let's do this like a freshman physics lab. This is one point of data in what will be a long term drift chart, with a mean, standard deviation, a number of samples and hopefully a time and date. Now regularly, say every week or every month, do the same. And maybe after a number of months you'll be able to say something about the relative long term drift, analogous to how measuring variation in distance between two space ships will tell you relative velocity but not velocity relative to any other body. Unless the relative velocity is really high, you'll need to measure over a long period of time because even with a linear relative drift of 10 uV/V/year, the daily drift would only be 0.02 uV/V.

...I kept on going, but didn't see anything more than .13ppm of 10V on any integration I observed. It may wonder on the high end of a .01ppm variation from 10V (meaning, not on the 9.9999999V side), and then creep up to wondering around a .1ppm variation of the same. In that sense, that's a "high" variation. It's typically more stable than this on a given evening, but I blame the time they both were off with its moisture accrual, wet caps getting a bit wacky, etc., and also just general environmental conditions fluctuations in an unconditioned space.

This is not data however.

Something to realize, is that running voltage references 24/7 creates a higher but more predictable drift. While switching them off reduces aging, but every time you switch a reference off and on there's a small chance it will jump in value. That's why Fluke 731/732 voltage references have a backup battery and an "in cal" light that goes off if the battery ever goes flat. Because only by maintaining power continuously between calibration cycles can they guarantee it will stay within specifications. Multimeters and multi-function calibrators generally have looser specs in the order of 1-10 uV/V/year, so this is less of a problem for them.