Products > Test Equipment
All about Keithley DMM7510. Bugs and features, recipes, advice, notes.
jjoonathan:
Speaking of linearity, what's the cal-lab technique for achieving it? I've heard that the 3458A has a special unique autocal self-linearization process that makes it 50x more linear than, say, the DMM7510, but what's the process? A really good Kelvin Varley divider or something?
TiN:
--- Quote from: E-Design on January 05, 2021, 07:22:58 pm ---Per Request,
Typical linearity performance of DMM7510. NOTE: data for one sided only. Other side is mostly symmetrical.
--- End quote ---
Excellent dataset, thank you. But it brings some more questions, as shape is bit strange if its indeed symmetrical.
I suspect that might be a linear fitting artifact, or there is really -0.2ppm pole at near 0V?
Also if there are datasets on other DCV ranges, that's be interesting. Some other meters are not as good for non-base 10(20V) range and INL for those rarely tested by people.
I also join to a question if you can share bit more about methodology?
Kleinstein:
The INL curve looks like relative to a linear fit for only the positive readings. Separate curves make absolute sense if something like Fluke 5700 calibrator is used to set the test points, as the fluke switches sign by relays and may show a small jump at zero.
From the data it looks like there is an additional meter used.
With a linear fit to only the shown data, the curve makes absolute sense. It looks like a combination of U² and U³ parts and some small more local wiggles from something like idle tones. A fit to the whole range including the negative side would better decide between U² and U³ parts. The absolute error would get larger, as the curve needs to join at zero. So the error bound do not directly apply.
U³ and U² parts can arise from resistor self heating and the FET switch resistance, so they are the normal errors to expect.
Especially the U³ part from resistor self heating may vary between units, as it is about proportional to the relative TC.
The curve still looks good.
MegaVolt:
A little about the good :)))
The device has a very low noise level. With inputs shorted in the 0.1V range and NPLC = 1, 42nV RMS can be obtained.
But if you add together several buffers with measurements and average them, you can reduce the noise level by the root of N times !!!
For example, here is the result of averaging 100 buffers.
RMS noise = 4nV !!!!!
P.S. such processing of measurements helps to reveal some measurement artifacts. I'll talk about this a little later.
Kleinstein:
I know about the usual reduction of noise level from averaging, but this only work so easy if the readings are independent, and this can be a problem with the 7510 for 2 reasons.
One is the ADC part, likely from the auto zero mode, there seem to be some correlation between the readings, e.g. from averaging the zero readings. This is one of the problems discussed in this thread for th noise in the 10 V range. There averaging does not follow simple theory. For the 100 mV range the ADC may not be that relevant, at least not the noise level for 1 PLC.
The second point is the configuration of the input stage. From the photos and analogy with other keithley meters the input section is expected to use low noise JFETs and an AZ OP to compensate drift. This would be very low noise at high frequency, but only moderately good at longer integration or averaging multiple readings in a row. To reach such a low noise level it would need more like classical auto zero, like in most HP meters with switching before the amplifier. So the AZ OP may be just there for the non AZ mode :-// :-DD.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version