Peeked over the precipice...

[EC8010]

As has been noted elsewhere, someone with one meter knows what the voltage is, yet with two you don't. Obviously, that means the more the better. I have six 6-digit DMMs, and three 7-digit. But which (if any) is right? Last year, I carefully tested them all at 10V from an (uncalibrated) Fluke 341A voltage calibrator. Interestingly, the most expensive meter (the 34470A) was the one furthest from the mean. Two 7-digit meters bought at a fair and claimed to be recently calibrated against a 3458A agreed closely and were close to the mean. In June, last year, I bought an ADR1001 evaluation board, which has been powered continuously since arrival and is intended to be a local check. Over the last few years, I've developed a low noise voltage reference (not LTZ1000 or ADR1000), and I'm about to make four of them (needed for other, less accurate projects). But four would be enough to test as an array to see how they drift relative to one another. I'm confident that my references are low noise, but I have no evidence to suggest that their long-term stability would be good. Even if there was low drift between four, that could just mean that they track ambient temperature nicely. Given that my meters/environment seem to be good for about 20ppm and no better, I can't make an absolute measurement.

In short, it looks as though 20ppm is my limit. My lab has a large South-facing wall, so its temperature oscillates like a yo-yo and I can't justify the expense of local temperature control. Improving from 20ppm would be a serious rabbit hole/precipice, and I have backed away; 20ppm is good enough. After all, most experiments are relative, looking for trends and correlations.

My question to others is: "Have you formally decided on accuracy limits?" If so, how, and why?

[Zenith]

What you trying to do?

I can see why fundamental definitions are necessary for mass, length and time, and how there were practical standards which were developed and refined. At one time lumps of platinum-iridium were the practical definition of the kilogram, but these were found to have problems and the practical standard had to be revised. Voltage is a derived unit and the practical (not very practical) standard was the Weston Standard Cell.
They were expensive to make and maintain properly. They were made with ultra pure materials, such as triple distilled mercury. They were also electrically delicate and drawing even 1 µA would degrade them. Places like the NPL would have banks of them, carefully temperature controlled and monitored. It must have been an expensive and cumbersome business. I don't know what companies like AVO and Simpson used for factory calibration. Maybe mercury cells. These days Josephson junction devices are the standard which defines the Volt. I believe you can buy one for £250,000.

So it's always been expensive and a full time job to maintain and use absolute standards.

Some amateurs are fascinated by metrology - time nuts and volt nuts. I can see the appeal, but it's a rabbit hole I have no intention of entering. I believe it gets very expensive.

My question to others is: "Have you formally decided on accuracy limits?" If so, how, and why?

No. I can't see the need. I have several 4½ digit DMMs which are all in close agreement and a couple of 6½ digit DMMs which are in agreement and I don't do anything that requires better than 4½ digits. It would nice to set them up against a much better standard, but not enough to pay to do it. I have a few WSCs but they are something of a curiosity.

As for time, I have a GPSDO which I assume is good to 1 in 10^12 when locked. That's essentially perfect for anything I want to cailbrate.

[tggzzz]

My question to others is: "Have you formally decided on accuracy limits?" If so, how, and why?

For me it is a case of how can I get excessive accuracy on a minuscule budget[1]. I use that as a way into the real pleasure for me: understanding all the ingenuity that has been required in metrology over the past couple of hundred years.

It all started in 6th form, using a 2% moving coil meter to measure voltages to 0.1%. It continues with meters that will measure 1kV and take zero current; none of this 10Mohm rubbish

As for need; rarely more than 3.5 digits, but I use my Agilent 34410A regularly, e.g. to measure stddev/noise. If I wanted to publish any results, of course I'd reconsider that.

[1] self imposed; I wouldn't get more pleasure if I spent ten times more.

[EC8010]

"What are you trying to do?" Good question. It all started out as a hobby... Then I was exposed to much better test equipment at work and an arms race between home and work benches started. But it's now a bit more than a hobby, and I habitually run computer-controlled experiments taking data from a number of instruments. Once I've wrapped everything up, I'll be looking to publish, which means my measurements need to be good.

I think I've decided that six-digit DMMs are as far as it is worth going; better accuracy demands a temperature-controlled environment. With a six digit DMM I can rely on 0.01% accuracy from a single instrument. (Somewhat simplified, but reasonable.) That means I can plausibly fit a model to measurements and a standard error of 0.1% is not marred by the instruments. That's good enough. There will be times when ratios can be taken with greater accuracy. But I don't need an 8.5 digit DMM...

What I have found in all this is that it is quite tricky to get real-world measurements with uncertainties that are as good as the instrument's specifications. Good test jigs are crucial and leads that came with an instrument are usually best ignored. Cables need to be kept short, and my best jigs hang off the front of the instrument's BNCs or 4mm sockets; they often include a temperature sensor.

[tggzzz]

That's all very sane.

It takes (rare) skill and imagination to ensure a measurement is valid. It takes (mere) money to have more digits.

[Zenith]

If you are contemplating publishing a paper on this you must have an idea of what the requirements are for the instruments used. It must depend on the size of the effect compared to likely instrument errors and noise. What would you expect to see if this was published by someone else? What would peer reviewers expect to ensure the journal wasn't publishing a load of nonsense based on inadequate measurements? They'd also look into the analysis and other aspects of the paper.

What I have found in all this is that it is quite tricky to get real-world measurements with uncertainties that are as good as the instrument's specifications. Good test jigs are crucial and leads that came with an instrument are usually best ignored. Cables need to be kept short, and my best jigs hang off the front of the instrument's BNCs or 4mm sockets; they often include a temperature sensor.

The people who develop the specifications for 6½ digit and better DMMs are experts who work with these things every day in dedicated labs. They won't be deceitful, but they will be out to present the product in the best light and they may take a lot for granted. Temperature, thermocouple effects, voltaic effects..... I can imagine everything getting critical when working with those standards of accuracy and precision, not to mention expensive.

[EC8010]

If you are contemplating publishing a paper on this you must have an idea of what the requirements are for the instruments used. It must depend on the size of the effect compared to likely instrument errors and noise. What would you expect to see if this was published by someone else? What would peer reviewers expect to ensure the journal wasn't publishing a load of nonsense based on inadequate measurements? They'd also look into the analysis and other aspects of the paper.

Agreed. Yes, it's a dynamic range issue. Is the noise/uncertainty sufficiently small compared to the parameter being measured? Typically, I'm measuring single-digit nA, or less. But my last job required me to learn how to measure 2fA to a reasonable accuracy, so I now consider 1nA to be quite a large current!

I think my measurement quality is good enough (actually, it's significantly better than any other measurements in the chosen field). And my analytical technique will pass muster. What I've learned with a vengeance is that measurement instruments and their jigs are like cameras and lenses. It's the quality of the jig that is crucial, not the camera body. Once that's done, money spent on better-than-six-digit meters is wasted unless the lab environment is controlled. I just wondered where other people had drawn the line...

Thermocouple effects only just come into six-digit DMM measurements (unless you're extremely careless). Oddly, the thing that really causes trouble is hum. And RFI from nasty supplies to laptops. I have learned to be very careful about both issues.

I tried the Lidl whisky you like; I hated it. Not to my taste at all. Give me advance warning of an event and I'll pass it on to you.

[tggzzz]

But my last job required me to learn how to measure 2fA to a reasonable accuracy, ...

Digital electronics

[MED6753]

My needs are much simpler. Two AD584-M references and an hp 3456A that probably hasn't been calibrated this century.

[mansaxel]

If you need more than 3 1/2 digits, you have done a marginal design

Or, you're trying science the real way.

I've got a couple 4 1/2 digits handhelds, and a 5 1/2 bench DMM (3478A) where the former owner bodged a battery swap. If I score the 6209a I'm ogling a the moment, I'll make an effort into getting at least some cal data into the 3478; it requires a 300V DC source to do, and the 6209a, while not calibrated per se can be adjusted to output that using my collective of 4 1/2s -- who all agree with my DMMCheck plus. Thereabouts lie my limits. And I hardly need them.

On lab space, my workshop is not temperature stable. Cold but not freezing in winter and quite hot in summer. There's a further limit.

Time, and frequency, OTOH. At work we have a requirement for 1µs precision on network time. I often, then, find myself in situations where this kind of precision is done. So, by habit and exposure more of a frequency nut than a volt nut.

[EC8010]

If you need more than 3 1/2 digits, you have done a marginal design

Or, you're trying science the real way.

I'd agree with that. What I'm doing at the moment is more physics than electronics.

[Zenith]

I'd say you have a serious application, and much more of a clue as to how to use these precision DMMs up to their limits than most of us, who in comparison like to do bit of mucking about. Hum and mains borne noise can be an irritating problem.

Thanks for the offer of the whisky. Probably the next time we are likely to run into each other will be at the Newbury or Chippenham rallies, around June next year.

[EC8010]

Hopefully, Newbury won't be cancelled this time; I always look forward to that one. I can't see when the Chippenham one is due.

[Zenith]

Chippenham is 25th July 2026 a Saturday.

https://chippenhamradio.club/rally/

Newbury is 5th July 2026.

https://www.nadars.org.uk/rally.asp

Newbury was cancelled last year because of weekend roadworks that were sprung on them at the last minute. Very much a one-off I'd say.

Andover - postponed last year, but not unpostponed - is provisionally 26th April 2026. I was worried that one might have disappeared.

https://www.arac.org.uk/events.html

Can't find anything about the Mendips bash, and the Mendips Repeater Group website doesn't mention it. I suppose it will be on.


Generally it's as well to look at the club website if they have one. The lists tend to be updated rather late and they can miss things altogether.