Bench DMM Noise & Stability Comparisons with TestController

[KungFuJosh]

After repairing my Keithley 2010, I'm not yet impressed with its stability. Though it isn't bad, I don't believe it matches its datasheet specs, so I will be adjusting it (TC=0 Zener diode stability) further before I decide whether or not to experiment with the internal reference.

This made me curious to see what other people are getting for stability with their bench meters. All are welcome to play, but I'm especially interested to see other K2010s, DMM7510s, 34470A, and similar meters. I'd also love to see how other 6.5 digit meters perform, but don't be shy if you have a lower digit meter and want to compare it also.

Let's see how our bench meters perform when comparing stability at different test lengths. Some datasheets claim a specific stability at 2 minutes, so I'd like to compare that along with longer periods. Let's say 2 minutes, 10 minutes, 30 minutes, or any number of hours. It's fine to screen capture the same test run over a period of time. However, if you wish to restart the chart and Grid panel readouts for each test, that shouldn't make a big difference if enough warmup time has already passed.

If you don't already have @HKJ's awesome free program TestController, get it!

To perform these tests, there should be substantial warmup time. Let's say a minimum of 4 or 5 hours (including the meters and voltage reference), but more is better. Because I want to rule out any amount of warmup drift, my meters have been warmed up for over 12 hours.

A quality voltage source is required to rule out other sources of noise/instability. For my tests, I'm using a battery-powered VREF10-001 r9 (it appears to be slightly more stable with batteries compared to the external supply). If you have something similar or better, fantastic. Absolute accuracy isn't important; it's the stability that matters in this case.

To chart these comparisons, we should use TestController, and make sure to use similar settings. I will add a post showing some preferred setup options. Ideally, this will include a chart, along with a "Grid panel" and "advanced readouts" for each meter (this is most important). That will show peak-to-peak (PP) and standard deviation (SD). If you're running a very old version of TestController I recommend upgrading before performing these tests. Feel free to include data exports (be sure to pick simple CSV or TSV so it can imported into Excel), but I'm most interested in seeing screenshots with grid panels for quick info comparison.

To get started, I'm sharing a couple 2-minute tests, and a 30-minute test. I'll update this post with more as time elapses.

Meter settings:
Range: Manual 10V (or 20V of your meter doesn't have a 10V range).
10NPLC.
1 sample.
No filtering or averaging.
Auto Zero on.
Line sync on if available.

TestController settings:
Grid panel with advanced readouts (see below for setup)
Chart with logging set to 1 second

Thanks,
Josh

PS. Noting environmental variables might be helpful too, especially if they're unstable. My office is currently stable at 24C (air is off).

[KungFuJosh]

For those less experienced with TestController, here is how you can setup your test and grid panels, and save it for quick repeat test setups in the future.

First, click Popups on the main Commands tab:


Choose Grid panel:


To control the Grid panel, you need to right-click on things. For example, where it says Grid layout, you can set that to 1x2 if you’re only using two meters:


Then right-click in the main area, and choose Advanced readout (for two devices, you must do it two times- once for each readout):


Then you have to right-click each readout and choose a channel:


I like to be fancy and change the colors:


I’ll usually do something like this, and try and make the panels match the chart colors (red device 1, blue device 2, etc.):

Note that you will need to select the meters in order on the Chart screen to match the colors you select.


This would be really annoying to do all the time, but you can do this once for your test setup and then save it as a menu (right-click the main Commands window):

That save feature is awesome, but keep in mind, it saves almost everything (you can edit that in the text file it saves though if you want to change something). For example, it will save the scales for chart, window positions, specific devices in use, colors of the popups, measurement type(s), etc.


To load your saved layout/test setup, right-click on the main Commands window again and choose setup:


This is setup is important because the main point of this test is to compare PPs in the readout popups:


Thanks,
Josh

[Kleinstein]

There is already a long thread on DMM noise comparison. That is however largely with a shorted input.
A difficulty in doing a test with applied voltage is having a low noise external reference. One can not easy tell appart what is noise / drift of the meter and how much is from the external reference. The ultimate performance, especially over longer time also depends on the temperature stability in the room.

[KungFuJosh]

All good points, thanks. For this though, I prefer to use a 10V reference, and mentioned using one of good quality is needed.

I think it's also useful for other members that have multiple meters and compare them at the same time. Even if the absolute values/stability may vary, you can see which meter performs better in each environment.

I agree, noting environment conditions would also be helpful, especially if somebody is only testing a single meter.

Thanks,
Josh

[HKJ]

If you want the curves more smooth, add a rolling average.

[Rubo]

Once I'm back from another trip to AZ next week, I'll finally finish unpacking the rest of my TE and after a couple of days of warm-up, will be able to add DM7276, 34465A, 34410A and ADCMT 7461A(maybe) to the list.
Since DM7276 can log the temp, it will be included too.

P.S. It's strange that according to the data you posted, your (not so favorite) 34465A IS the most stable, followed by Siglent in close second.

[Performa01]

Even though I own quite a colltection of bench DMMs up to Keithley 2001, I'm not set up for doing such tests anytime soon. Yet I've found an old excel workbook with stabiity tests on all ranges of my (heavily redesigned and improved) DC calibrator Analogic/Data Precision 8200A, Attached is the +10V measurement over a 10 minute interval.

For these measurements I've used the Picotest M3500A, which is exactly the same as a Keithley 2100. The advertising back then said something like "7.5 digit stability in a 6.5 digit meter".

So what you see is the combined stability of DMM and calibrator at an average 22.7°C. Sorry, humidity wasn't documented, but probably doesn't affect the 10 V range anyway.

[KungFuJosh]

Once I'm back from another trip to AZ next week, I'll finally finish unpacking the rest of my TE and after a couple of days of warm-up, will be able to add DM7276, 34465A, 34410A and ADCMT 7461A(maybe) to the list.
Since DM7276 can log the temp, it will be included too.

That would be great, thanks. Please make sure to include the popups with PP/SD from TestController.

P.S. It's strange that according to the data you posted, your (not so favorite) 34465A IS the most stable, followed by Siglent in close second.

It's not that strange. It is a great meter. It doesn't always win most stable, but for short-term tests it usually does.

The reason it's not my favorite is because I prefer UI of the DAQ, Keithley's superior support, and the higher level of control. I wrote a slightly longer comparison opinion here.
Most recent firmware:
Keithley DAQ/DMM: April 2026
Keysight 344XXA: August 2021
Siglent SDM3000X: April 2023

In no circumstances do any of them not need updates.

[Rubo]

Please make sure to include the popups with PP/SD from TestController.

Will do.

The reason it's not my favorite is because I prefer UI of the DAQ, Keithley's superior support, and the higher level of control.

I actually prefer the interface of 3446*A series, it's simpler. For what I do, it's easier to use. But it all depends on your use case, so I completley understand where you are coming from.

Up until about 2 years ago (last time I contacted Keysight) I must say that their support has been good. They always responded to all of my questions, sold me parts and were willing to calibrate/service all of my equipment (as a private individual, not a company). I know things have changed recently, but I haven't talked to them in 2 years, so can't say one way or another if it is different right now.

The only time I've had to deal with Keithley/Fluke was when I needed a replacement battery door for my 289 and they had no issues with selling it to me.

As far as firmware updates, any piece of equipment can/may/will need improvements. But any update needs to be tested (thoroughly), so you don't break something while fixing something else. Too many times have I seen an update f**k up something that was working fine before. So not sure if that's why there have been no updates to the meter in 5 years. Or if they are just not interested in it anymore???

[KungFuJosh]

I actually prefer the interface of 3446*A series, it's simpler. For what I do, it's easier to use. But it all depends on your use case, so I completley understand where you are coming from.

To each their own. Like I said, it's a great reliable meter (if you don't have one of the pale bastards).

Up until about 2 years ago (last time I contacted Keysight) I must say that their support has been good. They always responded to all of my questions, sold me parts and were willing to calibrate/service all of my equipment (as a private individual, not a company). I know things have changed recently, but I haven't talked to them in 2 years, so can't say one way or another if it is different right now.

They're all slow in communication, but for me, Keysight has been more cumbersome. They're all happy to sell you stuff or overcharge for calibration.

The only time I've had to deal with Keithley/Fluke was when I needed a replacement battery door for my 289 and they had no issues with selling it to me.

As far as firmware updates, any piece of equipment can/may/will need improvements. But any update needs to be tested (thoroughly), so you don't break something while fixing something else. Too many times have I seen an update f**k up something that was working fine before. So not sure if that's why there have been no updates to the meter in 5 years. Or if they are just not interested in it anymore???

Keysight just doesn't care. They were too busy making this crap to improve their advanced meters.

[Rubo]

To each their own. Like I said, it's a great reliable meter (if you don't have one of the pale bastards).

Mine IS one of those. Every time after it's been off for a while, I'm very "nervous" turning it on 

Keysight just doesn't care. They were too busy making this crap to improve their advanced meters.

Yeah, those never made sense to me. Granted, they have a VERY big display and with age I may appreciate it more, but as of right now, they just take up more space on the bench/shelf in my opinion.

[KungFuJosh]

 

[KungFuJosh]

I replaced VR122 (TC=0 diode) and R359 (sets current to VR122) in my K2010. You can see it was effective at improving stability and removing the related errors.

Thanks,
Josh

[KungFuJosh]

@TylerPeppy asked me my test setup for this, so here it is.

I have my VREF10-001 r9 battery powered, and sitting on a shelf, (hopefully) out of the way of the mini split:



All meters are connected with the same stackable banana plugs (yes, I'm aware better cables exist, thanks ).


I manually setup each meter to match consistent settings:
HighZ Mode (or match 10M instead - all should be the same)
Manual 10V Range (required because sometimes meters will be stupid and go up to a higher / less stable range)
Filter OFF
AZ On
Null/Rel OFF
NPLC 10
Sample Count 1 (any other averaging off)




Thanks,
Josh

[KungFuJosh]

These are the tests I ran last night with those settings. The K2010 has a freshly installed ADR1399KHZ in it now.

Thanks,
Josh

[KungFuJosh]

Here are some more updates. The ADR1399KHZ has improved a little in the K2010 as it ages.

I now also have an SDM4065A that was previously run 24/7 for at least 4 months by @TylerPeppy.

For comparison, I added in the SDM3065X. The SDM4065A was running for over 3 hours prior to that. Even without warmup, the SDM3065X outperformed the 4065A. With the 4065A running for over 7 hours, and the 3065X running around 4 hours, the 4065A is slightly worse, and the 3065X even better.

Test sequence first file timestamps:
Test 1 @ 2:30pm.
Test 2 @ 2:48pm.
Test 3 @ 4:29pm.
Test 4 @ 5:17pm.
Test 5 @ 5:41pm. (SDM3065X powered on with no warmup at this point)
Test 6 @ 6:44pm.
Test 7 @ 7:24pm.
Test 8 @ 9:33pm.
Test 9 @ 9:37pm.
Test 10 @ 9:40pm.
Test 11 @ 9:40pm.
Test 12 @ 9:44pm.
Test 13 @ 9:44pm.

Before warmup, the SDM4065A was typically around 11.9uVpp for the 2-minute stability tests. It jumped to around 17.9uVpp after a few minutes. The subsequent 2-minute tests did not improve over 7 hours.

Note: Only 2-minute tests attached to this post. Not all tests started with 2 minutes...I wasn't always paying enough attention.

Thanks,
Josh

[KungFuJosh]

Here are the rest of the screenshots with longer durations from the above test numbers.

Thanks,
Josh

[JohnRob]

Curious, how are you controlling contact thermal differences (material caused)?

[KungFuJosh]

Curious, how are you controlling contact thermal differences (material caused)?

All things are treated equally good or bad. I'm using basic stackable banana plug cables of equal length (I have a bunch of the same cables). They're all stacked in random order into the VREF10-001 r9.

I would prefer to use nicer shielded cables with better connectors, but that's expensive. It might improve the overall results slightly, but it won't do anything to even the already even playing field.

Thanks,
Josh

[KungFuJosh]

I let these run overnight. The results are kinda interesting. It looks like Siglent did improve the thermal stability of the SDM4000A series. The SDM3000X was the only meter more sensitive to the room temperature change.

That being said, it's not great to have a 7C temp shift if you're doing anything mildly important.

Also note that the K2010 case is not closed, and the rear bracket is not installed. So it's not surprising it's sensitive to room temp change.

I'd say it's pretty clear that while the SDM4000A is the noisiest, it's more temperature stable than the SDM3000X.

Thanks,
Josh

[Kleinstein]

The sensitivity to temperature changes can vary between different units of the same type of meter. With just 1 meter per type one can not say much about the meter types - it is still about the units at hand.
It is also not clear in how much and how fast the meters react to the temperature change. This may still need 1 or 2 hours more time, especially with the K2010, that has no fan and this makes it react rather slow. The open case may speed thing up. There can still be reactions to the rate of change.

All the meters still show an OK TC, like less than 1 ppm/K change.
With a room temperature change, there can also be a change in the reference voltage.

The way the K2010 is running one can expect a rather low drift with temperature, as much is corrected in "real time" (like 5 min scale).
In contrast the KS34461 relies on extra resistor ratios that can add more TC and drift correction via ACAL can be done by explicite cals, not in the background.

From the pictures show of the SDM3065 and SDM4065 - there is essentially no change in the analog hardware part and thus not mich difference expected. The SDM4065 still looks more noisy - so make sure the same settings are used (e.g. AZ mode, sampling rate). In theory there could be differences in the ADC setting and similar changes in the software that can effect the noise and drift.
I doubt it, but there could be a real time gain correction like with the K2010 - it would reduce the drift at the cost of more noise. Another option could be a correction from the measured temperature.

[KungFuJosh]

The sensitivity to temperature changes can vary between different units of the same type of meter. With just 1 meter per type one can not say much about the meter types - it is still about the units at hand.
It is also not clear in how much and how fast the meters react to the temperature change. This may still need 1 or 2 hours more time, especially with the K2010, that has no fan and this makes it react rather slow. The open case may speed thing up. There can still be reactions to the rate of change.

One way or another, they all show a temperature response related to the DAQ/PT1000's readings. I believe the open case on the K2010 does speed things up, but I'll eventually close it up and see how it compares.

All the meters still show an OK TC, like less than 1 ppm/K change.
With a room temperature change, there can also be a change in the reference voltage.

The temperature probe is placed on top of the voltage reference. There could certainly be a shift there, and that is somewhat likely considering the jump in Vpp across the board. In the future, I might setup a test with the voltage ref in a smaller temp controlled chamber to eliminate that.

The way the K2010 is running one can expect a rather low drift with temperature, as much is corrected in "real time" (like 5 min scale).
In contrast the KS34461 relies on extra resistor ratios that can add more TC and drift correction via ACAL can be done by explicite cals, not in the background.

From the pictures show of the SDM3065 and SDM4065 - there is essentially no change in the analog hardware part and thus not mich difference expected. The SDM4065 still looks more noisy - so make sure the same settings are used (e.g. AZ mode, sampling rate). In theory there could be differences in the ADC setting and similar changes in the software that can effect the noise and drift.
I doubt it, but there could be a real time gain correction like with the K2010 - it would reduce the drift at the cost of more noise. Another option could be a correction from the measured temperature.

So you think there's either a real time gain correction, or temperature compensation? That could make sense if you don't think there's a meaningful hardware change. The SDM4000A is both noisier and more temperature stable compared to the SDM3000X.

It could also be the case design has an effect on temperature stability, but that doesn't explain the higher noise level.

The attached test is over 21 hours of warmup time on the SDM4000A. It's worse (but consistent now) compared to the earlier tests that were consistent around 11.9mVpp. The results are about the same now as they were after 4 hours of warmup.

I wonder if the noise level can be improved via firmware if there's something dumb correcting gain or temp stability.

Thanks,
Josh

[KungFuJosh]

Here's my final series of tests for this run. Over 24 hours warmup on the SDM4065A. Only the SDM3065X has a few hours less.

The SDM4065A is more stable across environment changes, but consistently noisier than the SDM3065X. Even at 30 seconds (or less) the SDM4065A has substantial noise by comparison.

For comparison, the Agilent 34410A had similar stability / noise results over 90 minutes compared to the SDM4065A's 2 minute tests. Both K2700s tested better than that.

None of this is terrible, but I expect better from an allegedly "pro" DMM.

Thanks,
Josh

[TylerPeppy]

Interesting that the SDM4065A has an SD much higher than the others. I saw Kleinstein had mentionned the internal circuitry hasn't changed too heavily between the two, which can be cross-referenced with images of the mainboard for both SDM4065A and SDM3065X.

What is interesting, is how stable the SDM4065A is across the temperature changes, which leads me to wonder about the temperature compensation mechanism in the machine. I am wondering if some internal thermocouple is agressively compensating temperature changes...

How do the fan speeds on the SDM3065X and SDM4065A compare?

[KungFuJosh]

Kleinstein mentioned a couple options that could improve thermal stability, at least one of which could cause more noise.

As I mentioned, the design of the case can also improve thermal stability. I've not compared that between the meters either.

I have no idea regarding fan speeds, or how that's handled in them. Neither are notable regarding fan noise. Others more sensitive might disagree, but most meters seem quiet when you own a Hioki...anything. 🤣

I'm still tempted to upgrade the LM399, but the noise doesn't seem like it's just the ref at play. 🤷 If I do, I'll go the 10uF Tantalum cap route (I selected one with 5Ω ESR) so I don't need to add a resistor for the ref change, unless there's a convenient resistor in place already, but I doubt that's the case. Realistically, unless I get bored, I don't think I would change it for a while anyway.

Thanks,
Josh