DMM Noise comparison testing project

[bsw_m]

Amplifier is little bit noisy, but have very low input bias current (~0.07pA).

  What transistor is at the input?  How are the protection circuits made?

As an input differential pair, a custom hybrid transistor assembly with full custom transitors die (developed and produced in MNIPI) US006 is used, containing 2 differential jfet pairs.
The absolute maximum input current for it is no more than 3pA
Dual jfet 2PS104 are used as protective diodes with low leakage.
some time ago I uploaded a photo of the schematic diagram to the cloud storage to share it on Russian-language forums: https://disk.yandex.ru/i/SuMbouTaVEutcA

[Kleinstein]

The circuit looks strange for modern times - still using a photoresistor chopper, at least allready with LEDs and no neons. less than 1 pA bias suggests that there is some adjustment/compensation of the current of some kind (should be possible through the LDR chopper).

[bsw_m]

You defeated me.

Oops  Have you measured the input bias current?

Keysight B2985A noise.

2pA range, 100NPLC, input closed with dust cap.

V7-49
Uncontrolled environment.
Filter on.
2pA range, input closed by cap.
About 13hours data collection.

Due raw data file is too large to attach. I'm upload this to cloud disk: https://disk.yandex.ru/d/QeBi84xEmjFU2A

[bsw_m]

The circuit looks strange for modern times - still using a photoresistor chopper, at least allready with LEDs and no neons. less than 1 pA bias suggests that there is some adjustment/compensation of the current of some kind (should be possible through the LDR chopper).

This DMM start development in late 80, and production start in about 1991-1992 year.
photoresistor chopper - developed in MNIPI for low termal EMF, low noise and some other things. Named OR46
Compensation for bias current not used/present, this can be seen in the schematic diagram.

Hmm.... may be I need start topic for this DMM?   

[dietert1]

Another noise test of a Keithley 2182A nanovoltmeter:
 
K2182A settings:
:SYST:AZER? => "1"
:SYST:FAZ? => "1"
:SYST:LSYN? => "1"
CH1, 10 mV range
1 PLC
240V AC setting with 220 V synthetic mains supply (DDS generator + power amplifier + transformer)
Meter is inside cardbox
Internal input short (solder bridge)

Logging every 5 seconds. On average 25 samples + int temperature reading take 4.256 secs per point.
Since the K2182A will be used with a low thermal mux, both analog and digital filters are off.

Result:
Clean 3 hour log with 1.4 nV rms standard deviation over 2160 points!
bottom up noise = average of standard deviation of 25 samples per point / 5
top down noise = standard deviation of point to point differences / sqrt(2)

Regards, Dieter

Edit: A step from 220 => 210 V AC resulted in a zero shift of about 23 nV. Half of that happened within minutes, the other half took 1.5 hours. Probably isolated power supply first and then mains transformer heat-up. In total about 5 nV for each % of mains voltage deviation. My synthetic mains setup is good for about 0.1 %. A second K2182A exhibits the same problem.

[_pv]

without autozero

data: https://github.com/pavel212/noise

[Kleinstein]

The curves for the 3458 and 2182 DMMs show quite some dips at 20 ms and 40 ms integration. This suggests that there is quite some residual hum in the signal, and not just noise.

[_pv]

The curves for the 3458 and 2182 DMMs show quite some dips at 20 ms and 40 ms integration. This suggests that there is quite some residual hum in the signal, and not just noise.

inputs are just directly shortened with 0 Ohm, what else could i do.
all the questions to voltmeter manufacturers why PSRR is that bad, so it is visible with shorted inputs when aperture time is not an integer number of PLC.

[maxwell3e10]

Thanks for this great summary plot and welcome to the message board!
What does NMV stand for?

[_pv]

NMV stands for a typo in NVM:
https://github.com/jaromir-sukuba/nvm/tree/main/measurement_data

[MegaVolt]

The curves for the 3458 and 2182 DMMs show quite some dips at 20 ms and 40 ms integration. This suggests that there is quite some residual hum in the signal, and not just noise.

Keithley 7510 shows the same behavior.

https://www.eevblog.com/forum/metrology/dmm-adc-noise-comparison-testing-project/msg3115162/#msg3115162

[Kleinstein]

The curves for the 3458 and 2182 DMMs show quite some dips at 20 ms and 40 ms integration. This suggests that there is quite some residual hum in the signal, and not just noise.

inputs are just directly shortened with 0 Ohm, what else could i do.
all the questions to voltmeter manufacturers why PSRR is that bad, so it is visible with shorted inputs when aperture time is not an integer number of PLC.

The amplitude goes down at 20 ms, but not much at 10 ms integration. So the background is more 50 Hz hum and not 100 Hz. I would suspect magnetic coupling, either from the DMM internal transformer or other transformers or similar sources of a magnetic field. External fields could be shielded with some steel or just reduced with more distange, or a less susceptible position. For the 3458 the connection at the guard terminal can also make a different. Chances are it should be connected to COM for the noise test.

[_pv]

The amplitude goes down at 20 ms, but not much at 10 ms integration.I would suspect magnetic coupling

yes, for poor PSRR rectified 100Hz should be present, but
for magnetic coupling in order to induce some microvolt * 10ms on a some few cm^2 loops require magnetic fields more than earth field, which i'm pretty sure i do not have in the lab where some of these noise plots were measured.

[Kleinstein]

Looking at the numbers an external magnetic field is unlikely. 1 cm² area and 1 µV hum would need some 33 µT, which is unlikely, though still within permitted safety limit of 100 µT. More typical fields in the lab should be less than 1 µT. The 3458 circuit is physical quite large and may have a larger effective area, but hardly that large, unless they made a big misstake (e.g. have some magnetic material inside the area to amplify the effect / concentrate the field).

There is still the meter internal stray field from the transformer. This can be more a problem with 50 Hz than 60 Hz as the magnetization would usually run higher. With transformers primary designed for 60 Hz operation this can become a problem, especially with more audible hum at 50 Hz.


For the 3458 the question is, if the guard is connected to low and maybe also to ground. This can effect capacitive coupling.

[_pv]

For 3458A it was just a "floating" shortened input without guard connected to LO button.
And if it is not PSRR, then it is awful CMRR, as it is still 0 Ohm shortened input, and you could not inject much onto a 0 Ohm through capacitive coupling differentially.
And for common mode there is also no tesla coil or similar stuff around with huge electric fields, especially at 50Hz.

Keithley do not have guard terminal at all, just a diff input, and behave the same.

[Kleinstein]

The protection for the voltage input has some resistance (10 K for the 3458) in series. So the actual amplifier does not see a dead short but these 10 K. The 3458 has a switch to connect the guard to lo for a reason. The protection in the 2182 should be rather low resistance, so I would not expect capacitive coupling to be an issue.

The visible about 1 µV of 50 Hz hum is also not very much. In most cases the extra suppression from integration over 1 or 10 PLC would deal with this. The extra CMRR from integration over mains periods is in the 60 dB range and better (depending on how good the mains frequency is). It is only with the shorter integration time that it really matters. They have the gurad terminal and a switch to connect to low if wanted to improve on the CMRR.

[alanambrose]

Do these look right? i.e. is the K2182A much better than the 34420A re noise?

Alan

[maxwell3e10]

The problem with K2182A is that it doesn't integrate well for longer times. The 0.01V scale noise never really goes below 1 ppm=10nV. In contrast, 34420 noise improves as 1/sqrt(t) as it should, and crosses 1 nV at a time of a few seconds.

[dietert1]

Above you will find how i obtained 1.4 nVrms for 5 second integration time with the K2182A. I used it's built in data buffer and averaged 40 samples. That was without any mods. So I'd guess the two intruments aren't so much different. Rather it is very easy at that level to introduce thermal EMF, e.g. in the front panel connector.
In the thread about multiplexers for monitoring voltage references i posted some results with the K2182A. They show how to make the instrument work stable to 1 nV over a week or so with the help of a relay multiplexer.

Regards, Dieter

[Kleinstein]

The data for the 2182 look suspeciously good at 1 PLC and not that good for longer integration. There may be some hidden digital filtering or maybe not using simple AZ mode.
The Keithley 2182 gets extra complication from having separate auto zero settings for the ADC and input amplifier and some extra analog filtering. So it really depends on the settings and in some cases there is extra settling time. The NPLC setting alone can be misleading, even without extra digital filtering. The case with input chopping, but no AZ mode for the ADC is also a bit misleading, allowing good performance over a short time, but some 1/f noise and drift over the long run.
Working well at 1 PLC and not so good with longer integration is a problem with many Keithley meters - not just the 2182. My suspision is some hidden digital filtering that improves on the 1PLC noise at the costs of extra noise at 100 PLC.
The ADC in the K2182 has a potential for good noise, possibly approaching the K2002 and K2010 noise performance, as these are not that much different. In parts they seem to compromise INL for noise.

For shorter integration the 34420 is limited by the ADC: it is the same ADC as in the 34401 and for less than 10 PLC (and maybe even at 10 PLC) limited by quantization noise. It is kind of a very good input amplifier, but a relatively noisy ADC, especially at 1 PLC, but also the 10 PLC ADC performance is not great.

[bsw_m]

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

[dietert1]

What is the data rate? Do you know how to calculate the standard deviation?

Regards, Dieter

[bsw_m]

What is the data rate?

Data rate is 3sps.

Do you know how to calculate the standard deviation?

Stdev=1.2nV

Regards, Alexey

[alanambrose]

>>> The data for the 2182 look suspeciously good at 1 PLC and not that good for longer integration. There may be some hidden digital filtering or maybe not using simple AZ mode.

The noise specs below do seem to suggest that the marketing folks have picked only the NPLC=1 case to demonstrate noise results with no filter.

I'll check my code and maybe run again - it certainly looks like it always selects AZ mode.

>>>
he ADC in the K2182 has a potential for good noise, possibly approaching the K2002 and K2010 noise performance, as these are not that much different. In parts they seem to compromise INL for noise.

For shorter integration the 34420 is limited by the ADC: it is the same ADC as in the 34401 and for less than 10 PLC (and maybe even at 10 PLC) limited by quantization noise. It is kind of a very good input amplifier, but a relatively noisy ADC, especially at 1 PLC, but also the 10 PLC ADC performance is not great.
<<<

Ah, interesting insights.

Alan

[Kleinstein]

The noise specs aready include the relatively high noise for the slower cases. So it is not only the 1 PLC case and some of the more disappointing case are included. Some of the numbers look odd, like only about a factor of 2 (or even less for the 100 mV / 1 V range) improvement (instead of square root of 10) for the step from 1 PLC to 10 PLC. One the other hand the step from 18 PLC  to 50 PLC helps more than is should in the 1 V range.  Chances are the 600 nV for 18 PLC in the 1 V range is wrong as channel 2 got better specs in this case.

The measured noise for the 10 V range and 1 PLC is close to the specs - so the ADC really seems to be that good.


Giving the settling time to a varying noise level is a bit strange. The settling may also depend on the range, as it looks like the lower ranges include some analog filtering. Chances are the 10 V range settles faster than others.

There are seprate specs for the reading speed and these indicate that the DMM is spending quite some extra time: e.g.  15 readings per second for 1 PLC (50Hz) instead of some 24/s as expected for a simple 2 conversions AZ cycle. There is something else going on, like additional zero readings or reference readings for the scale factor like in the K19x).  For actual use the reading rate can be the more important parameter than the actual integration time.