DMM Noise comparison testing project

[Kleinstein]

For sensitive experiments, that are disturbed by cosmic radiation one can do a parallel measurement of the radiation. This is nothing new - I know it is done in some experiment, dropping data from times with high radiation background. The cosmic background is not very smooth but to a large part comes in burst. So for short intervals of something like 10 µs you get bursts of high radiation, with reasonable long time in between with much lower level. So if these high radiation times can be identified, it can be better to ignore this part of the data. There are also variations from the sun, that can be so intense that even plane avoid some regions (near the magnetic north and south poles) at some times.

Though I don't think the input stages of typical DMMs are that sensitive - this might be different when measuring in the fA range.

For the input stages of a DMM or small amplifiers I would more think of the influence of single electronic states that are either occupied or not - at effective positions these can give measurable jumps. Also the position of single dislocations can have an effect.

[alanambrose]

This is a x-post from plesa's 34420a repair thread, but it should be interesting here - a first cut at a 34420a noise floor with an OK but not-perfect short...i.e. a Lemo connector with some copper wire - but not soldered in yet with the specified Sn/Ag solder.

Any 'turning up at the 200 nplc points' might be temperature drift (or just statistical variation - RSE is targeted at 10%). My meter is reporting firmware versions: 9.0-5.0-2.0 btw.

Curiously channel 2 has much lower noise. Whether this is my measurement error or a bad channel 1 or real results, I'm not fully sure - so I would be very interested in someone else verifying this result on their own meter. Looking at the schematic, the channels are asymmetrical (e.g. channel 1 can source current) so it is just possible that the channels have a very different spec. The docs are actually a bit brief on the specified noise floor which is a bit disappointing.

Anyway, I hope that's interesting.

[splin]

This is a x-post from plesa's 34420a repair thread, but it should be interesting here - a first cut at a 34420a noise floor with an OK but not-perfect short...i.e. a Lemo connector with some copper wire - but not soldered in yet with the specified Sn/Ag solder.

Any 'turning up at the 200 nplc points' might be temperature drift (or just statistical variation - RSE is targeted at 10%). My meter is reporting firmware versions: 9.0-5.0-2.0 btw.

Curiously channel 2 has much lower noise. Whether this is my measurement error or a bad channel 1 or real results, I'm not fully sure - so I would be very interested in someone else verifying this result on their own meter. Looking at the schematic, the channels are asymmetrical (e.g. channel 1 can source current) so it is just possible that the channels have a very different spec. The docs are actually a bit brief on the specified noise floor which is a bit disappointing.

Anyway, I hope that's interesting.

A small point but are you sure about the 34420A 10V noise specs (the brown triangles)? The 10V range data sheet spec is 450nVrms for "6.5 digits (10 PLC) with Analog Filter Off and Digital Filter Medium (50 reading average)." Surely that is equivalent to 500 PLC rather than the 10 PLC shown in the graphs?

Also the legends for the blue dashed (green in 2nd graph) '344020A noise spec' should actually read '1mV range, typical' - though there doesn't seem to be much difference between them; datasheet spec is 1.3nVrms @ 500 PLC whereas the typical (from page 63 of user guide) is 1nV.

[plesa]

I know that there is a difference between channel-1 and channel-2 --- one of them can go up to 100V and the other one only goes up to 10V, but I can't remember which is which.  So, one of the channels is hooked up to a HV divider and possibly some different protection network that might explain the difference in noise performance.  What does the manual say about this?  My 34420A is in my home laboratory and I'm not there right now, but I will try to test this later tonight.

Channel 1 is max 120V rated and channel 2 is max 12V rated. On my repaired unit the channel 1 is little bit more noisier than channel 2.
It takes some time to settle down after changing the channel.

[zlymex]

, I'm seeing something that can only be described as "popcorn noise"-- little positive [or negative] "spikes" in the data that far exceed the input bias current-- seemingly random, but at a far lower rate than the more-or-less normal input noise. 

Is it something similar to my measurement as attached?
This is my femto-amp meter(DIY), the CMOS input with a very large feedback resistor. I cannot find the source of those spikes neither. Once I even suspect it was the cosmic ray or background radiation because I measure the similar background in my Geiger counter.

Yes, very similar, but my "spikes" are much larger.  I also thought that it might be cosmic radiation--- huge bundles of mixed particles and rays at incredibly high energy levels---  it would certainly explain the data perfectly.  It would be instructive to connect a recorder to the DMM output, and input a highly divided signal from a radiation detector to the DMM---  that way, you would see a "jump" in the signal with each event from the detector, and you would also be able to see if the spikes coincided with the radiation data ["spike" on top of the "jump"].  OR, if you have multiple DMMs, then record the data from them all simultaneously over GPIB, and then plot the data from each so that you can see if the spikes coincide on two or more meters.

If this is cosmic radiation, well then there is little we would be able to do about it other than a sophisticated filter designed to pass the "normal" data while blocking the "spikes".  That would be material for a master's degree thesis...

I came across similar "spikes" several times when I measure a 10V or 7V with my 3458A. Strangely, those spikes were all downwards. This not only happened to me, also happened to others thousand miles away. The reason could be interference from the mains or from the air.

As for the femto current meter, I made another one to test them at the same time, the spikes were not synchronized.

[zlymex]

Not cosmic radiation then-- you would see it in both channels.  Might be radioactive particles in the packaging, but I thought they solved that decades ago...

These are very low current levels, so maybe just shifts in the local field getting picked up?  [People walking by, truck goes by in street, etc.]  At these levels, to be very honest, we should be using shielded and guarded measurements and special cables [like triax, with the middle shield a driven guard, etc.].  Circuits should have careful consideration to shielding and guarding, etc.  Most DMMs don't have this [the 3458A does, but few people make use of it].

Also maybe just quantum mechanics getting us [possibly "hot" electrons being randomly generated in silicon lattice imperfections, with some chips better than others]...  Don't know...  Beyond my knowledge at this point...

Well, the current sources, current-voltage converters, and handheld voltmeters(with data recording ) are all battery powered. The current sources and current-voltage converters are all in their own aluminium cases. I put all of them into a large aluminium box with 4 holes but I cover those holes by coins. Thru one of the holes I can see the voltmeters reading. So I believe what ever the noise is, must come from inside. The earth connection was not shown because I moved the box to the floor for photo.

[alanambrose]

>>> A small point but are you sure about the 34420A 10V noise specs (the brown triangles)? The 10V range data sheet spec is 450nVrms for "6.5 digits (10 PLC) with Analog Filter Off and Digital Filter Medium (50 reading average)." Surely that is equivalent to 500 PLC rather than the 10 PLC shown in the graphs?

Hmmm, that's an interesting thought - I took that to mean that the noise was calculated over 50 readings - but maybe it was ~50 readings @ 500nplc equivalent.

>>> Also the legends for the blue dashed (green in 2nd graph) '344020A noise spec' should actually read '1mV range, typical' - though there doesn't seem to be much difference between them; datasheet spec is 1.3nVrms @ 500 PLC whereas the typical (from page 63 of user guide) is 1nV.

Yes you're right re '1mV range, typical' - I see that if I take the 10nplc figures from the datasheet and the 'typical' behaviour of page 61 of the user manual I should be able to plot ~noise floor for each range.

BTW I should have a general purpose test executable debugged over the next few days for others to test with.

Alan

[zlymex]

Many thanks for the explanation re ref noise and other noise contributors to the acquisition process - I am beginning to understand now. Also an interesting point about fourier analysis. Below a plot comparing azero on and off (I used the same setting on all nplcs) to counteract drift - presumably the difference is mainly down to temperature drift. With a coupla runs I was able to get an internal temp drift less than 0.1C during the nplc 1,000 run. These are at 60 samples per nplc for a relative standard error in the noise readings of 10% (see error bars). TiN - it would be interesting to run the 'azero on' test on your 'evil' meter - maybe your temperature environment is fairly drifty like mine.

(edit) fixed the error bars on the chart and added the intervening points.

Alan

This is a very good chart
Did you measure a 10V voltage reference or the input shorted? And what that 01704 means on the title?

[plesa]

Alan provided me his test program so we have direct comparison. Measures on bench in my lab without any additional enclosures with 34103A short.

[Kleinstein]

There must be something very wrong with the noise curves for the 34420. It does not make sense for the lower ranges to show lower relative noise. Except for the higher ranges that use the divider, the noise relative to full scale should go up when going to higher amplification (if limited by the input amplifier, which is expected for the low ranges) or stay constant if limited by the ADC itself (may be the case for the 10 V and 1 V range of the 34420). So it seems to be more the absolute noise that is plotted.

I also doubt the 34420 with it's rather simple ADC could be better or even close to the 3458 in the 10 V range. It may be in the 100 mV range due to the better amplifier and less protection.

[alanambrose]

>>> Did you measure a 10V voltage reference or the input shorted? And what that 01704 means on the title?

This is with a not-great short rather than 10V - a couple of ordinary Pomona shorting plugs if I remember, not the low thermal ones.  '01704' is the last part of the serial #.

>>> There must be something very wrong with the noise curves for the 34420. It does not make sense for the lower ranges to show lower relative noise.

Yeah there's some add behavior here - I can't see why the channels are so different either. That's why I'm interested in some independent verification. I guess it's possible to use the built-it stats functions to check these results...

Regards, Alan

[Kleinstein]

The 34420 specs are for 50 the average of 50 conversions and only a rather limited time frame (only 2 minutes with only 5 independent readings per minute) - so the specs mark should be for effectively 500 PLC and slightly (e.g. 5-10%) corrected up for the too short frame.  The 10 PLC value would than extected to be about 7.5 times higher than marked, thus more like 3-4 µV_RMS für 10 PLC in the 10 V range.

Most of the measured curves for the 34420 are also somehow  not plausible - more like using numbers for RMS noise in µV - but the scale shown in ppm of range !. This at least would be about at the specs.

The data for the second channel like all the same - like using one of the lowest ranges for all curves. Maybe setting the range did not work with channel 2 ?

From the schematics the two channels do not look much different - so not much different noise expected.There are possibly slightly different paths for different gain setting and also the amplifier can change. Chances are that the 10 V range noise is due to the ADC

[Cerebus]

These are very low current levels, so maybe just shifts in the local field getting picked up?  [People walking by, truck goes by in street, etc.]  At these levels, to be very honest, we should be using shielded and guarded measurements and special cables [like triax, with the middle shield a driven guard, etc.].  Circuits should have careful consideration to shielding and guarding, etc.  Most DMMs don't have this [the 3458A does, but few people make use of it].

Also maybe just quantum mechanics getting us [possibly "hot" electrons being randomly generated in silicon lattice imperfections, with some chips better than others]...  Don't know...  Beyond my knowledge at this point...

To put this into perspective, 100 fA is 624,151 electrons per second (i.e. less than 1 electron per microsecond). At those rates it doesn't take too many electrons to upset the apple cart. I don't trust my maths to work out a theoretical noise current at these levels but I know it won't look good; whatever, once you take basic statistics and quantum effects (this is Fermi-Dirac territory) the shot noise alone is going to be significant simply because the arrival of one single extra electron in a bit more than a microsecond doubles your current.

To look at it another way, you're not measuring DC any more, you're measuring a 624kHz square wave with terrible phase noise and significant random amplitude modulation. :-)

More prosaically, I think the noise being seen is probably good old popcorn noise. Some level of popcorn noise is unavoidable and at these levels what would normally be insignificant noise suddenly looks like a big event.

[Alex Nikitin]

At last I had a chance to measure noise on the HP3458A (Option 002) in my work lab. Here is the result - looks quite close to specs?

Cheers

Alex

[TiN]

That's what I see on my box too.

[alanambrose]

Re 34420a - apologies I have lots of parallel projects running v slowly

I would be very happy, of course, for someone to use some alternate technology: ezgpib / labview / the built-in stats etc ... to come up with some comparison numbers.

Also I would be interested to see the numbers for the Keithley nanovoltmeter.

AoE, of course, has a lot of description of the 34420A front end.

Regards, Alan

[TiN]

I need to get around measuring mine 182M.

use some alternate technology

Alternate to what? Pi+linux-gpib? Numbers should be same, disregarding of data transfer method, I'd expect.

[alanambrose]

OK these are my 34420a results (well for 6 data points) punching the buttons on the front and using the meter's own built-in stats functions:

duh

Methodology (if anyone wants to replicate with their own meter) was:

0 apply the short
1 power-up into default settings
2 set filters off
3 set the channel and range
4 set the nplc in the MEAS/INTEGRATE menu
5 set on the NULL and then STATS functions (toggle to clear the NULL/STATS)
5a wait for 60+ samples
6 read out the stats with Shift >
7 GOTO 3 (or 4)

[Kleinstein]

The new data / table for the 34420 make much more sense.
It's interesting that the 1 PLC reading at 0.1 V range still seems to be limited by the ADC, not the amplifier. So a very good amplifier, but a not that good ADC, especially at high speed - about what one expects from a low level meter.

[Jorn]

The results for Keysight 34470A. Measured directly on the instrument using build in instrument statistics. Rising curve from 10 to 100 NPLC on 100mV range must be due to input circuit drift...

Sample count as follows:

NPLC	#samples
0.02	100000
0.06	30000
0.2	10000
1	2000
10	200
100	50

Using to many samples on the 100 NPLC measurement results in drift contribution however using to few increases uncertainty on stats  

Hope anyone can confirm the results. Quite satisfied with 10V - 10 NPLC result

-jorn
(Posting his very first message in the eevblog forum)

[alanambrose]

Ah thanks Jorn for the 34470A results, very interesting.

Hey Kleinstein re:

>>> It's interesting that the 1 PLC reading at 0.1 V range still seems to be limited by the ADC, not the amplifier. So a very good amplifier, but a not that good ADC, especially at high speed - about what one expects from a low level meter.

Could you explain how you made that deduction - are you working backwards from the theoretical noise floors of the amplifier and adc?

TIA, Alan

[Kleinstein]

There are four main noise sources in a DMM:
1) the ADC itself: this gives a constant noise contribution relative to the range.
2) The reference noise: this only applies if a voltage other than 0 is measured - so it does not matter for data on shorted input.
    It is proportional to the measured voltage.
3) The noise from the input Amplifier. This noise depends on the range / amplification chose. Usually a constant noise voltage and thus increasingly important in the low voltage ranges.
4) Noise from the input protection / divider: High resistance in the input protection (e.g. series resistance) gives noise, that behaves similar to the amplifier noise. A special case is the typical 9.9 M - 100 K divider resistance, that can contribute to noise especially in the 100 V range.

The noise relative to the full scale in the 0.1 V range at 1 PLC is even lower than in the 10 V range, which is a little strange, but might come from the special kind of range switching used. It's also possible to have quite some quantization noise at 1 PLC and thus a noise level that can change with offsets / the exact measured voltage.

If there would be significant amplifier noise, the 0.1 V range should show a much higher noise than the 10 V range. So the noise must be mainly the range independent ADC noise. It's also typical to have very little amplifier noise for the range without extra amplification (e.g. 10 V range).

A second factor that point towards noise from the ADC is that the noise at 1 PLC is much higher than at 20 PLC. White noise from the amplifier would make the noise go down with the square root of the integration time. With a contribution of 1/f noise the noise would go down even slower with longer time. In contrast to this the ADC can have a noise contribution (e.g quantization or determination of residual charge) that goes down inverse proportional with the integration time - this about what is found here for the 34420. So that ADC is much better at 10/20 PLC than at 1 PLC.

[alanambrose]

Ah v interesting, thanks for taking the time to explain that.

Alan

[Jorn]

Below follows the detailed report for Keysight 34470A Noise floor measurements. Data capture is automated using a small python script communicating through GPIB. (The 34470A build in statistics has limited resolution and is somewhat tedious to utilise for an extensive analysis of the noise floor.)

All measurements are performed with auto zero on. Time unit on x-axis is power line cycles.
This means that the x-axis spans from 2s to 2000s(!). Multiply by 2 to get actual acquisition time using azero.

The first point for all NPLC100 measurements is absent because it consists of only one sample.

I think the graph's indicates that statistics settles around 2000-5000 PLC and that drift kicks in from 5000-10000 PLC.

This leads to the conclusion that noise floor measurements can be tricky to carry out and that the results has to be used with caution in comparisons with other instruments...
 
-jorn

[TiN]

That is interesting angle and good details on your test.
I'd like to try that python script, it should be easy to adopt it for other meters to perform equal comparison.