R&S NGM output readback: Resolution vs Accuracy

[jusaca]

I'm currently looking at the datasheet of the Rohde & Schwarz NGM power supply series. In general they seem like quite some capable instruments


But I'm a bit confused by the specs of the readback accuray:

The readback resolution is really great, 1ppm is nice. But the accuracy is factor 1000 worse, only 1 permille of the range as base accuracy!
So e.g. in the 100mA range we get 100nA resolution (awesome!), but only 100µA accuracy (not so awesome...).

Does that make any sense? Why offer a resolution three orders of magnitude above the actual accuracy?

PS: Link to datasheet (specs on page 15):
datasheet: https://scdn.rohde-schwarz.com/ur/pws/dl_downloads/pdm/cl_brochures_and_datasheets/data_sheet/3609_1685_32/NGM200_dat_en_3609-1685-32_v0400.pdf

[squadchannel]

lab power supply shouldn't have that accuracy, and it isn't even necessary.

It's just an unnecessary digit added for the sake of showing off.
In fact, only three/four digits in the setting. so, only need three/four digits, right?

[tooki]

I'm currently looking at the datasheet of the Rohde & Schwarz NGM power supply series. In general they seem like quite some capable instruments
(Attachment Link)

But I'm a bit confused by the specs of the readback accuray:
(Attachment Link)
The readback resolution is really great, 1ppm is nice. But the accuracy is factor 1000 worse, only 1 permille of the range as base accuracy!
So e.g. in the 100mA range we get 100nA resolution (awesome!), but only 100µA accuracy (not so awesome...).

Does that make any sense? Why offer a resolution three orders of magnitude above the actual accuracy?

This isn’t actually that unusual in test gear, even if it usually isn’t as big a factor as here. The extra resolution still lets you see small changes, even if the absolute values might not be 100% correct.

[tooki]

lab power supply shouldn't have that accuracy, and it isn't even necessary.

It's just an unnecessary digit added for the sake of showing off.
In fact, only three/four digits in the setting. so, only need three/four digits, right?

Nonsense. Having the extra readback precision isn’t for showing off, it’s to make it unnecessary to have an extra piece of test equipment to measure small differences.

[jusaca]

Hm ok, I see.
I hoped I would be able to see directly on the supply whether my nRF board is drawing 1µA or 2µA in the power down states with different settings. But guess accuracy of a power supply (even the very good ones) is just not made for that. Still, would be very convenient

I still feel three orders of magnitude higher resolution than accuracy is slightly extreme and a bit misleading.

[squadchannel]

you'll need something like an ammeter or Dave's uCurrent.

lab power supply shouldn't have that accuracy, and it isn't even necessary.

It's just an unnecessary digit added for the sake of showing off.
In fact, only three/four digits in the setting. so, only need three/four digits, right?

Nonsense. Having the extra readback precision isn’t for showing off, it’s to make it unnecessary to have an extra piece of test equipment to measure small differences.

That's true. I can see the change.

[thm_w]

0.05% DC I is the same as a 34450A 5.5 digit meter: https://www.keysight.com/us/en/assets/7018-03637/data-sheets/5991-1133.pdf

Usually you can calibrate out the offsets and readback accuracy in terms of offsets is better than advertised. In this case I don't know if you can calibrate it yourself.
Though you'd need a 6.5 digit meter and at that point you can just use that to take the measurement.

[tautech]

Hm ok, I see.
I hoped I would be able to see directly on the supply whether my nRF board is drawing 1µA or 2µA in the power down states with different settings. But guess accuracy of a power supply (even the very good ones) is just not made for that. Still, would be very convenient

I still feel three orders of magnitude higher resolution than accuracy is slightly extreme and a bit misleading.

Not for a SMU it isn't. 

[Kleinstein]

With modern ADC chips high resolution is cheap. High accuracy still needs good shunts and a good voltage reference and not all ADC have a stable gain.
So they can get the high resoltion for cheap, but high accuracy would add significant costs, also for calibration and testing.
The voltage readback error is not that high, it is mainly the more tricky current part.

[jusaca]

Okey, I kind of get what you're all saying, but struggle a bit with understanding the exact benefit.
If the resolution is cheap to increase so much like here, it might not have a real impact on the price point of the instrument. But when all these additional digits are just guesswork, I don't really see the benefit.
Even seeing relative changes or trends is not all that useful if the absolute accuracy is so "bad" (I mean, bad in comparison to all these digits on screen) that the trend might be entirely wrong...?

[Kleinstein]

Especially with the shunts a large part of the uncertancy is from the shunt resistance that effects the gains.
Another part with shunts is the self heating effect: the shunt gets hot and change the value.This way the shunt performce is worse near the end of range and better over a good part of the range. This is especiall for the highest current range, less for lower current ranges.
So the last digit may not be valuable when near the full scale, but can still be ok when the current is only a small fraction of the range.

[tooki]

Okey, I kind of get what you're all saying, but struggle a bit with understanding the exact benefit.
If the resolution is cheap to increase so much like here, it might not have a real impact on the price point of the instrument. But when all these additional digits are just guesswork, I don't really see the benefit.
Even seeing relative changes or trends is not all that useful if the absolute accuracy is so "bad" (I mean, bad in comparison to all these digits on screen) that the trend might be entirely wrong...?

But it’s not guesswork; if the instrument is halfway reasonably designed, the extra digits may have an error, but it’s not random: the error comprised of gain errors (input x some value), offset errors (input + some value), and noise. The noise is fairly easy to deal with by averaging or other filtering, so let’s ignore it. So your display value = (input x gain error) + offset error.

Suppose the actual input value is 1000. Let’s add a gain error of, say, +1%, and an offset of 0.005.
Display value = (1000 x 1.01) + 0.005 = 1010 + 0.005 = 1010.005

Now let’s assume the input value changes to 1000.001.
(1000.001 x 1.01) + 0.005 = 1010.00601

There’s still an error and the absolute value is wrong, but the change of .001 was still visible. That might let you see things that are useful.