Analogue input accuracy calaulation method

[Ross]

Hi all,

Before I ask my question I want to apologies if this is a simple question already answered on this forum or on the web. I have been unable to find anything, probably due to not knowing the technical name for the process makes it harder to search for.

In specifications for devices that use ADC (say a DMM) the accuracy is given as %range +%measurement (can be expressed in different ways). I wonder how this is calculated?

Thanks Ross

[HalFET]

You mean how they calculate the possible error range when using a component, or how it is measured in the first place?

[Ross]

You mean how they calculate the possible error range when using a component, or how it is measured in the first place?

I was thinking that both would be part of the answer, but if not then I am more interested in the calculation part?

[beanflying]

In theory it is best done by calibration against known standards in a calibrated Lab or at least referenced against standards much more accurate than the meter.

It could also be done mathematically which is what you are thinking but without a specific circuit and all component values and complete data sheets this cant be done and it very complex for all but the simplest circuits.

I have for example just finished a couple of Voltage references and apart from 2 filter Caps, minimal wires and a small proto board there is just one Chip containing the Zener . So in this case providing you have minimised copper losses, noise (caps and power supply) and met the data sheet specs claiming this reference for example at X +- Y microV would be ok. BUT when you get to this level (lack of) Temperature Coefficients and short and long term drift are much larger than the initial spec. So a complete Spec would X +- Y +- Z/degC +- Qppm/time ......   This reference for example is way more accurate than my 3 1/2 and 4 1/2 digit Multimeters and more stable than my 6 1/2 digit which drifts more than it does. Confused yet 

What I suspect happens in the case of some manufacturers is they break out a dart board throw a bunch of darts and take an average 

[Ross]

In theory it is best done by calibration against known standards in a calibrated Lab or at least referenced against standards much more accurate than the meter.

Thanks fro your reply, just to be on the same page, when you say using a know standard, this is using feeding in a known accurate voltage into the ADC and the output is recorded?

I assume this would be done at different voltages over the range of ADC measurement then feed into some maths to get the %range + %reading? If I am correct what is the maths?

[nfmax]

The % of reading + % of range error specification comes from the implied model of the measurement error: a linear scale factor error plus a constant offset (on each range). The % values come from the maximum allowed values of each.

[beanflying]

In the case of multimeters it is the meters accuracy as a whole that is tested and not just the ADC. A lot of modern meters are then tweaked in software against a standard or known voltage, current or resistance etc. to improve their calibration. Old school they would have had trim pots in the circuitry to do the same thing.

Writing a meters %range and %reading also must take into account any non linearity of the overall meters design. In theory these figures are also a worst case figure on the meters produced so some will be well inside this spec and none outside.

Two of my meters below both 5 year old against a 5V Reference (4.99989 at 20 degrees this morning) meters turned on for about 5 minutes and the reference has been running for several days.

Victor 86B spec on DC 400mV/4V/40V/400V/1000V   ±(0.5%+4)

So for a 5V nominal input this could be 5 +- 0.025V+0.04 or 5 +- 0.065 both meters are well within their spec.

Dropping the same across a badly constructed voltage divider (measured 0.99957Vdc) of Vishay 0.005% resistors

For this nominal 1V input spec would be 1+-.005+.004 or 1+-0.009

once again the meter is well within spec. This would then be repeated for AC, Resistance and all the other ranges.

To arrive at this 'assumption' however I am relying on my Agilent 34401A being perfectly calibrated which I have no real way of knowing as it lapsed 2 years ago. Due to it's accuracy and precision I have a high degree of 'certainty' that the figures I used from it are more than up to the task of calibrating my Victor meters due to their lower accuracy and precision. If it was in calibration I would have a reference to a known standard and could actually put a number to that 'certainty'.

There is a thread running in the metrology section on the forum about getting started with some links to good reading.

[Ross]

The % of reading + % of range error specification comes from the implied model of the measurement error: a linear scale factor error plus a constant offset (on each range). The % values come from the maximum allowed values of each.

Thanks nfmax, I think I understand the definition of accuracy error and how to apply it from a given spec to obtain an error for a give measurement range. My questions is how these spec are obtained. For example if I am a DMM manufacture and I want obtain the error spec for the product. How is it calculated?

Apologies for my delayed response.

[Ross]

There is a thread running in the metrology section on the forum about getting started with some links to good reading.

Thanks for your reply and sorry for the delayed response. I cannot seem to find the thread you are taking about in the metrology, never been good at searching forums. Can you provide me a link or the title of the tread? Sorry to be a pain.

[tautech]

A little tip for expressing plus/minus in posts on this forum.

Use the + from your keyboard then highlight the + with your mouse and use the Underline button above.

Looks something like this before posting [ u ] + [ / u ] (spaces added)
Comes out like this: +

[CopperCone]

I think about the measurement you take being a error bar with the tolerances specified.

So I just imagine its fuzzy between this and that.

The error bar is also there to simply the span calculation.

No doubt, everything would benefit from some kind of complex polynomial with 100 measurement points. Look at the INL for a ADC. You could smooth that out, like the equation for RTDs (calsen van dusen or something like this).

You could have it bang on at 10v, but at 1V it might be off because of INL. They calculate it based on the tollerance of the components (worst case specified by manufacturer) and the INL of the ADC. The INL is small though, most likely.

If you get into the nitty gritty, for alot of things, single point calibrations suck unless the transfer function is known and a equation to match that function to the measurement is used. Most likely this would require significant computational power. 

And a "best fit" line might be difficult to generate. Probably would result in some beastly polynomial for random things.

not to mention some large number of sample points.

[beanflying]

Thanks for your reply and sorry for the delayed response. I cannot seem to find the thread you are taking about in the metrology, never been good at searching forums. Can you provide me a link or the title of the tread? Sorry to be a pain.

https://www.eevblog.com/forum/metrology/how-do-i-get-started-98964/

Getting to the more nerdy 6+ digit readings sees stated %accuracy very very low on equipment and Temperature Coefficients or TempCo drift cause a lot of the error or uncertainty in the readings. If you want to see hyper nerdish talk of that check out the LTZ1000 thread