DMM Noise comparison testing project

[_pv]

10hours noise & stability of 33years old V2-38 nanovoltmeter:
10uV range no filter no dumping. Plotted raw data.

could you please share raw data as well?

[bsw_m]

could you please share raw data as well?

No problem!
Measurements were made on the range of 10 μV
First column is time in unixtime format. Second column - data from nanovoltmeter.

[alanambrose]

Ah, two conversations going at once

Re Kleinstein's thoughts:

>>>  So it is not only the 1 PLC case and some of the more disappointing case are included.

As the only no-filter specs were 1 nplc, those were the only ones I used for comparison with my no-filter tests. I agree that the specs are confusing - maybe innocently, they're using some 'real world' examples. And I don't pretend to fully understand the noise specs because of the '2D' combination of nplc & filter data points. There's probably some rationale there but I don't quite see what it is. Is the response time really 'the time you realistically need to wait for the results to settle' rather than an input variable and should therefore really be one of the rightmost columns?

For the 34420A (below) I noted the official specs on the original graph for the 1mV range of 16 and 1.4 ppm at 1 & 200  nplc - although right now (the actual tests were done some time ago), I can't see where I got those from. However, I've also added the 100 nplc official specs on the graph (also below). All said, I'm thinking my K2182A & 34420A results roughly match the official specs and so my analysis is good. Any thoughts on that?

Alan

p.s. I still need to double check the two 2182A auto zero settings ('Front Autozero' and 'Autozero') I used in the test.

[Kleinstein]

The digital filter is simple averaging. So one has to multiply the PLC setting and fitler setting to get the total integration time. Many meters automatic convert PLC settings larger than 10 to real intgration of some 10 PLC and than use averaging as digital filtering.

[_pv]

Measurements were made on the range of 10 μV

Just updated the plot,
and "noise" here is StDev over 10 consecutive measurements at given averaging time

[alanambrose]

>>> The digital filter is simple averaging. So one has to multiply the PLC setting and fitler setting to get the total integration time. Many meters automatic convert PLC settings larger than 10 to real intgration of some 10 PLC and than use averaging as digital filtering.

OK understand, that makes perfect sense.

Alan

[n_haku]

Mine data from ad7461a. Though cal expired approx half year ago, it should not affect in adc zero noise measue. Data are srtaight, w/o any math, after well warmed, (>2h), rear input shorted with bare copper wire, every run 30 min lenght.
If matters, I can collect data for longer periods, when dmm not used.

PS: as there seems no dedicated thread for equipment's ROMs, I'll also attach dumps for tr6861, r6144, r6551. I checked function work before read.

[MegaVolt]

Mine data from ad7461a.

What is the y-axis dimension?

[n_haku]

What is the y-axis dimension?

Since no math applied, its in lsb of range, ie 10uV for 10V, 1uV for 1V, 0.1uV for 100mV, and cause max readings are 1e6, it become ppm of range.

[zrq]

Reviving this thread with some data taken from a Keithley 2001M with front input shorted.
It's noise is very disappointing:
1. almost a order of magnitude higher than modern 7.5 digits and some 6.5 digits DMMs
2. can hardly be averaged down at 100s scale (very significant 1/f)
3. 200 mV range is particularly noisy, not really providing higher resolution than the 2 V range in long averaging time.
I'd like to sell the meter and try to get something else (oh no, I don't want to be a voltnut and spend x5 the money for a DMM7510).

Here shows the allan deviation of various different settings, unit in Volt RMS and in ppm range for easier comparison.
The raw data is attached in the compressed files in the next post, first column is time in unix timestamp and second the voltage. A few traces have (low-resolution) room temperature recorded in a separated file. The BATT ones are measuring a 9V battery, not included in the allan deviation plot.

BTW, does anyone know if it's possible to completely disable the screen to save VFD life? The screen off command kept the VFD on with a fixed screen disabled message, increasing the sampling rate at 1PLC from 19.7Hz to 22.2Hz, completely useless.

Edit: added a few traces in the volt scale plot.

[zrq]

Selected raw data for Keithley 2001M, a few long and high speed recordings cannot be uploaded due to attachment size limit.

[bsw_m]

Noise and stability the R3003 microvoltmeter - part of R3003 voltage comparator.
Microvoltmeter in 10uV range. The output digitized by DMM 100PLC speed.

First column in file - time in UNIX time format. Second column - digitized result which is given to the used range of the microvoltmeter. No more mathematics was used.

[iMo]

Years back I did with my 34401A at 100mV/100NPLC/AZON with 42nV stddev/rms.
I want to repeat the measurement at .1V/1V/10V - what NPLC should I use? 10 or 100?

[Kleinstein]

There is somewhat limited need to have a run at 100 PLC, as this would averaging 10 PLC data internally. So a long enough rund at 10 PLC could be used as well. The Allan deviation plot should take care of the averaging already and the curves should overlap to a large part.
1 PLC may be interesting, but for the 34401 this has quite some noise from quantization. Still possibly a point for the 100 mV range as less 1/f noise is involved.

[iMo]

Ok, I will do die langen Runden with 10PLC (fast 6 digits)..

[zrq]

Recently, I observed another surprising behavior of the Keithley 2001M. When the option slot is empty, whether it's covered or not can have a significant effect on the zero point.
My meter comes with neither option card nor the slot cover, so I simply filled the port with some aluminum foil. The observation began with noticing a sudden jump in a long recording of 20V range, followed by another jump back close to the original value later in few thousands of seconds. Initially I believe it's a jump in temperature or something else, but a similar jump happened again when recording for 200V range and this time I'm sure everything is stable. Later, I noticed another similar jump when I was working on the GPIB cable and torn off the filling aluminum foil accidentally, and the value quickly restored after I put another piece of foil filling the option port!
For now, my best explanation of the observation is the internal temperature distribution can be slightly changed by the airflow. The aluminum foil can get loose and move a bit. So far, after I cover the option port with tape and foil more carefully, I haven't seen such jumps ever again.

[iMo]

Here is a 34401A measurement at 10PLC, 100mV range, AzOn, not fully warmed-up, raw data and put into Plotter app. Is this Overlapping Allan Dev the measure we are interested in? Hopefully my settings are ok..
Edit: nope, wrong tau0, fixed..

PS: at 4.04sec tau (100PLC) it shows 38nV which fits my 42nV@100PLC stddev measurement above..

[Kleinstein]

The Allan deviation is a useful form to show the result, as one can also see how good one could get with averaging. Still not everyone is familier with this format.
The simple standard deviation for a fixed window (the lenght 100 or 1000 points is debateable) is easier to understand. So one should give that number too.

The x axis should be in seconds. Normally 10 PLC should be close to 2.5 or 3 readings per second - so the data indicate 50 Hz mains frequency. I think the plot is wrong and need the sampling time and not the samplig rate entered as tau0, as the curve should start at some 0.4 seconds and not 2.5 seconds.
Not fully warmed up adds the drift, that could well be the reason for the upward terend after some 2000 units. So especially the low ranges should be well warmed up.

[iMo]

The Allan deviation is a useful form to show the result, as one can also see how good one could get with averaging..

Thus my above measurement says I can go for 5minutes with averaging at that range.
I messed with TimeLab a lot when playing with OCXOs, a pity the Plotter app cannot read csv files, I had to replace commas and rename the file..

[Kleinstein]

The Allan deviation curve again going up is likely due to the drift, not the intrinsic noise. With a a more stable temperature the drift should get smaller and thus even longer averaging may work. There will finally be a limit from temperature fluctuations or background signals (e.g. hum, EMI).  The noise level is already not bad, considering that much of the noise is due to the resistor string (8*13K) used for the protection and the balancing resistor in the zero link.

[iMo]

Ok, after warming up my 34401A here are - the 100mV and 10V MADEVs at 10NPLC and AZON over night runs and 1V measurement made over the day.
Needs some analysis on this method and results..

PS:
a) removed the Plotter outputs
b) added 10PLC 100mV, 1V and 10V MADEV in TimeLab

Note: The 1V range has been sampled during the day with more temperature drift around, will be repeated over night sometime..