4 wire & 2 wire resistance tests, various cheap Kelvin clips, various meters

[SilverSolder]

I would expect the repeatability to be much better than those specs, if the measurements are made within a few minutes of each other?

That is absolute measurement accuracy. Short term it should be well qirhin 24h specs for relative measurements.

Understood.  So it might just be possible to distinguish between very small values, as long as you do it fast! 

[Electro Fan]

Big Disclaimer/Request:  no laughing.... 

No doubt low value resistance measurements have a lot of potential variables including finding the right Kelvin clip test leads, and much more (temp, materials, settling, etc, etc).   But.... back to basics:

Given two 4 wire measurements, one showing .0092 ohms and one showing .0097 ohms, is the difference within the specs for the 34465A?  (I'm having a difficult time interpreting the specs.  Thx)

https://www.keysight.com/us/en/assets/7018-03846/data-sheets/5991-1983.pdf

Specifications 34465A
34461A accuracy specifications (pdf says it's for 34461A but I think it's a typo and is for 34465A as well as the 34461A and 34470A):
± (% of reading + % of range)

Relevant Spec?
100 Ω 0.0030 + 0.0030
100 Ω is the range?

So, if we start with .0092 ohms
then we multiply .0092 x .0030% and we get 0.000000276 ohms (not much?)
then we multiply .0030% x 100 ohms (the range) and we get .003 ohms (this is the big one?)
then we add it all up:

.000000276 + .0039476 = .003947876

This would indicate that for .0092 ohms the error could be +/- .003947876 ohms

If that is correct (and I'm betting it's not) that would indicate there is no chance of accurately distinguishing between .0092 and .0097 ohms.

Ok, feel free to issue my F grade for interpreting the specs.   

Thanks

The math seems correct, but I would use 1 year specs for it, which are 100 Ω : 0.0060% of value + 0.0040% of range ( unless your instrument was calibrated in last 24 hours..):

(0,0092*60e-6)+(100*40e-6) =0,000000552+ 0,004= 0,004000552 which gives 0,0092 +/- 0,004000552 Ohm as your interval..

Thanks for taking the time to review the math and for confirming it is correct for 24 hour measurements.  (I guess that means for measurements made of the same thing made within 24 hours?)

It looks like the one year measurements allow for greater tolerance - but even with the 24 hour measurements the spec seems to be saying that 9.2 milliohms could range from nearly 5.2 milliohms to nearly 13.2 to milliohms.  Based on my various experiments/tests including trying to reproduce the theoretical resistance for 45 cm of 18 gauge copper wire the meter and Kelvin clips seem to be doing much better than the spec.  Which leaves me wondering why the spec would seemingly have so much tolerance.  Maybe it's a general purpose DMM and Keysight doesn't want to go too far out on a limb for resistance measurements and/or maybe they would prefer to sell a specialized milliohm or micro ohm instrument. 

In any event, it's not like anything I am building needs to have 9.2 vs 9.3 milliohms - but after figuring out some of the testing variants (including accounting for temp, settling, and the fairly small Kelvin clip resistance) I'm happy to see that the measurements and trends are starting to make sense and that the meter does seem to significantly exceed the spec.  Mostly it's been a good trip a few feet into the rabbit hole and I'm inclined to climb back out for the time being.   

I've attached a couple more screen captures showing the 45 cm wire over 16 plus hours (and it's now two hours later (~18 hours) and the average is still .000949 so I'm pretty sure right or wrong on the exact value the meter and clips have reached a pretty stable reading and it's a reading that the theory and math say is pretty reasonably close to what should be expected.

[Shock]

Check out this other 34465A thread and the below video. It shows a guarded measurement workaround on the 34465A. If you are diving into micro ohms especially with long cables you need to take a holistic approach. If it was me I would start off with a 2 and 4/5 wire shorting adapter just evaluate the meters settings after running through the auto calibrate and then null etc. If this is done before jumping into different cables and performing measurements you can have a baseline from which to evaluate introduced "measurement noise" then build up from there. Which is why baselines are so important, anything you add including changes in the test environment will have a cumulative effect on the results.

As a consideration if you are squeezing out performance for low ohms compare spending money on expensive cabling against a LOM 510A. One (for repair) was sold here a few weeks back for $75 which seemed a bargain as they are $2000 instruments. This is the route I went a while back, it becomes a lot easier to evaluate cabling, connections and surfaces. You hit the wall pretty fast in low ohms on most instruments.

[Electro Fan]

Check out this other 34465A thread and the below video. It shows a guarded measurement workaround on the 34465A. If you are diving into micro ohms especially with long cables you need to take a holistic approach. If it was me I would start off with a 2 and 4/5 wire shorting adapter just evaluate the meters settings after running through the auto calibrate and then null etc. If this is done before jumping into different cables and performing measurements you can have a baseline from which to evaluate introduced "measurement noise" then build up from there. Which is why baselines are so important, anything you add including changes in the test environment will have a cumulative effect on the results.

As a consideration if you are squeezing out performance for low ohms compare spending money on expensive cabling against a LOM 510A. One (for repair) was sold here a few weeks back for $75 which seemed a bargain as they are $2000 instruments. This is the route I went a while back, it becomes a lot easier to evaluate cabling, connections and surfaces. You hit the wall pretty fast in low ohms on most instruments.

That is an excellent video.  What it shows is consistent with what I've experienced (without having the guard). 

The LOM 510A is a very cool instrument but at some point either space or the budget (or both) could be stressed.

Most importantly, the path you describe above (highlighted in colors) makes a ton of sense.  Thanks for the insight, advice, and encouragement.

[Electro Fan]

Last post on this particular test before shutting down the test.

At 25 hours it looks like it continues to be be very stable with an average of 9.49 milliohms.

[Electro Fan]

Ok, I couldn’t "resist".   

Found my new favorite Kelvin leads...

Here is another test using a set of BK Precision TLDK1 Kelvin leads.  You can compare the day long results to the post above; also attached is a third image showing an interim short term (one minute) snapshot - I think it indicates a pretty steady state that probably mostly reflects the ADC trying to do it's thing but that might also reflect some fairly stable characteristic of the BK Precision Kelvin clips and leads.

Overall the test shows similar performance to the Kelvin leads I posted before (in the post above).  The BK Precision leads averaged 9.46 milli ohms vs. the other (eBay) leads that averaged 9.49 milli ohms.  Both tests were about 25 hours.  My guesstimates, based on my 45 cm 18 gauge copper reference wire, indicate that the BK Precision’s lower measurement (9.46) is closer to what is expected based on theoretical analysis.  Also, the BK lead measurements seemed to dance less on the trend chart with a max 1 day variation of 0.8 milli ohms vs a max 1.5 milli ohms 1 day variation with the eBay leads.

Having said that, I think the difference in the 9.46 vs 9.49 measurements is possibly within the technique used for null or offset compensation (I’m still trying to get a grip on null vs offset comp) - but so far if I had to recommend one or the other I’d go with the BK Precision TLDK1.  The materials and construction look somewhat better on the TLDK1 and given that BK Precision recommends the TLDK1 with their 2840 and 2841 resistance meters, I think they are probably worth the extra $30 (~$50 vs ~$20).  I could be wrong but I don’t think either the BK or eBay leads are shielded so that’s another step up to be addressed if you need/want that.  One advantage of the eBay leads would be that if you had to put the two Kelvin clips apart from one another the eBay leads would reach further (approximately 2 meters vs 50 cm) from one end of a DUT to the other (doesn’t seem likely unless you are measuring wire, and theoretically that could perhaps be coiled up – not sure if coiling is optimum though; without coiling the wire I think the eBay leads will allow a 2 meter separation).

Here are some links for both Kelvin leads (and the BK Precision meters):

eBay “LCR Meter Cable w/ Banana Plug Connectors kelvin clip SMD” 4-Wire Kelvin Test Leads
https://www.ebay.com/itm/LCR-Meter-Cable-w-Banana-Plug-Connectors-kelvin-clip-SMD/381460680487?ssPageName=STRK%3AMEBIDX%3AIT&_trksid=p2057872.m2749.l2649
$19.95 plus shipping

BK Precision Model TLDK1 4-Wire Kelvin Test Leads
https://www.tequipment.net/BK/TLDK1/Kelvin-Test-Leads/?Source=googleshopping&gclid=EAIaIQobChMIubrx_Y326wIVDb7ACh3bAQs_EAQYAyABEgLdZ_D_BwE
Replacement test leads for the 2840 and 2841 DC Resistance Meters
MSRP $52.00 (on sale for $47.30 plus shipping)


https://bkpmedia.s3.amazonaws.com/downloads/datasheets/en-us/2840_Series_datasheet.pdf

https://www.bkprecision.com/products/component-testers/2840-dc-resistance-meter.html

https://www.bkprecision.com/products/component-testers/2841-dc-resistance-meter-with-temperature-correction.html
Model 2840 2841
Display Range 1 µΩ to 20 kΩ 0.1 µΩ to 100 MΩ
Basic Accuracy 0.1% 0.01%

Video demo of TLDK1 leads on a BK Precision meter and a 34465A

[Electro Fan]

Another test with BK Precision Kelvin TLDK1 Kelvin clip leads.

This test was about 40 hours.

Slightly different results.

30 micro ohms higher average (9.49 milliohms vs 9.46 milliohms) on the 40 hour test vs the 25 hour test.
Slightly less variation (9.2 - 9.9 milliohms over 40 hours vs 9.1 - 9.9 milliohms over 25 hours)

[Electro Fan]

Ran another test to see if the 34465A would stop at 50,000 readings.

After about 96 hours it's up to over 51k readings and still going, and the results are pretty consistent with the previous ~1-2 day tests.

[Electro Fan]

Still going.  After about a week (187 hours) the 34465A is still happily sampling.

I had interpreted the spec to say the base memory was sufficient to do 50k samples but after revisiting the spec it says 50k readings.  I would have thought that each reading occurs when the trigger fires (and that a reading was synonymous with a sample), but apparently the reading is just that (a reading) and a reading involves multiple samples (presumably tied to the PLC setting) so the sample readout is just that, the number of samples - not the number of readings.  If I was Keysight I might have chosen to display the number of readings (in addition to also showing the number of samples).  In any event at 100 PLC and 100k samples this thing is still going strong a week later and I don't envision testing anything for more than a week.

I'm happy with the 34465A and the BK Precision TLDK1 Kelvin leads.  The 45cm 18 gauge copper wire DUT is also performing admirably. 

[tooki]

Still going.  After about a week (187 hours) the 34465A is still happily sampling.

I had interpreted the spec to say the base memory was sufficient to do 50k samples but after revisiting the spec it says 50k readings.  I would have thought that each reading occurs when the trigger fires (and that a reading was synonymous with a sample), but apparently the reading is just that (a reading) and a reading involves multiple samples (presumably tied to the PLC setting) so the sample readout is just that, the number of samples - not the number of readings.  If I was Keysight I might have chosen to display the number of readings (in addition to also showing the number of samples).  In any event at 100 PLC and 100k samples this thing is still going strong a week later and I don't envision testing anything for more than a week.

Looking at the user manual, it really looks like they are using the words "reading" and "sample" almost interchangeably, and "measurement" occasionally as a synonym, too (in addition to "measurement" as the general concept of measuring). It looks like they just call it a "sample" in the context of the "digitizer" function, but that it refers to exactly the same thing as a regular reading. From what I glean from the manual, "regular" readings are taken by auto triggering (where an algorithm decides "oh, I think this is an actual reading and not noise") or from being manually triggered. The digitizer mode, in contrast, takes readings at regular intervals, meaning it's acting as a DAC, so they go with "sample" there.

But unlike with the terminology, the manual is very clear on how the memory works: it's a FIFO buffer, so it simply discards the oldest readings/samples/measurements once memory is full. So yeah, it's telling you it's taken 100K+ readings/samples/measurements, but only the newest 50K are actually being evaluated. You can likely let it go and take millions of readings/samples/measurements, but it doesn't mean the memory is actually holding them all.

There's an SCPI command to query for overflow once this happens.

[Shock]

Averaging over 20/40 hours is fairly meaningless anyway aside from the min max during that period.

[Electro Fan]

But unlike with the terminology, the manual is very clear on how the memory works: it's a FIFO buffer, so it simply discards the oldest readings/samples/measurements once memory is full. So yeah, it's telling you it's taken 100K+ readings/samples/measurements, but only the newest 50K are actually being evaluated. You can likely let it go and take millions of readings/samples/measurements, but it doesn't mean the memory is actually holding them all.

“but only the newest 50K are actually being evaluated.”

Any chance even though the memory is being replaced FIFO style the stats including min, max, and also average might be based on the full duration of the test?

[tooki]

That’s a good question, and I don’t know. It’d be easy enough to test, though.