4-Wire Kelvin probes

[todorp]

So I ordered some probes from Ebay and AliExpress and now am waiting for arrivial (will take some time). I started thinking of a test setup/procedure to be able to tell which probes are bad/ok/good. Of course there is build quality (materials), ruggedness, ergonomics, etc. But what I am interested in for the time being is the electrical performance. I have a HP34401A which hasn't been calibrated in a long while and a DE-5000 LCR. I plan to use the kelvin clips with both of these instruments.
The question is: what is a standard procedure that can be used to assess the quality of the 4 wire kelvin connection on each one of these instruments? For example do people use a standard 1% resistor (which value?) or something better and measure it? What parameters do you setup in the instrument (nulling, NPLC)? How long do you people measure for? What post-processing do you apply to the measured data?

Bump - no one?

[mawyatt]

Don't normally require highly accurate reading down in the sub-ohm region, and if we do then use the Kelvin Probes (TH26011CS) that came with the Tonghui TH2830 LCR meter.

With the cheap Kelvin clips mentioned, first is to get a good Zero Ohm reference. A short section of solid 12AWG Cu wire or thick piece of Cu plate/sheet acts as a good short reference, be sure to clean the Cu for a good contact, then "Null" this with the 34401A. We found this gives good results that compare well with the TH2830 in lower value resistors. One difficulty is getting the same location for the clips on leaded resistors when measuring low values, so try and position the clips in the same location for comparisons, with higher values resistors, this isn't a problem.

For reference resistors, one should have a quality standard. We don't, and just use our KS34465A as the "Reference" reading instrument to compare other instruments, components and probes. Not exactly "Metrology Standards", but works OK for our needs.

Awhile back we did design a custom PCB to host some precision low TC resistors that we can use for relative measurement comparisons, but this fixture plugs directly into the DMM without cables, has PCB mounted banana plugs. This uses header clips to set the resistor under measurement and arranged so the clips resistance doesn't influence the readings, and has a set of pins for external measurement. However, these pins will affect the external reading and partially negate the advantage of the external Kelvin Probes since the measurement current and sense now will both "see" a small series resistance due to the resistance setting header clips.

Think these cheap Kelvin clips are OK for GP use, and cheap enough to give a try, but don't recommend them for any precision low ohms work.

Hope this helps.

BTW for Precision Kelvin Tweezers for use with SMD chip components, the Tonghui TH26009B Test Fixture is excellent. These are very smooth and easy to use, hard to describe just how good they perform. They aren't expensive (~$120) but do have the 4 fixed BNC interface you find on precision LCR meters, so not compatible with a typical DMMs.

Best,

[Ground_Loop]

I'm giving these away for shipping cost, US only.  PM me if interested. The Pot and Cur leads connect to opposite sides of the clip jaws.

[Electro Fan]

So I ordered some probes from Ebay and AliExpress and now am waiting for arrivial (will take some time). I started thinking of a test setup/procedure to be able to tell which probes are bad/ok/good. Of course there is build quality (materials), ruggedness, ergonomics, etc. But what I am interested in for the time being is the electrical performance. I have a HP34401A which hasn't been calibrated in a long while and a DE-5000 LCR. I plan to use the kelvin clips with both of these instruments.
The question is: what is a standard procedure that can be used to assess the quality of the 4 wire kelvin connection on each one of these instruments? For example do people use a standard 1% resistor (which value?) or something better and measure it? What parameters do you setup in the instrument (nulling, NPLC)? How long do you people measure for? What post-processing do you apply to the measured data?

Bump - no one?

I adopted this approach to making a reference "DUT".  What I like about it is that you can research what the theoretical measurements should be for various gauges, lengths, and compositions of "copper" wire.  Then you can see how close you can get your clips, leads, and meter (working together as a system) to measuring the theoretical values.  If you look at enough web sites you might find that some web sites claim different specs for the same wire values - generally small differences, but something to stay on the lookout for.  Some web sites provide values and others have calculators - either way, watch for different claimed or assumed values.  By the time you dial in the supposed theoretical values you will then find that various other factors will influence the test results - such as how and where exactly you place the clips on the wire ends, the ambient temperature (including any fluctuations) in your test area, how long you let the test setup settle before beginning the tests, how long you run the test, your meter's ability to collect data over some test period (and your meter or your ability to average the samples), and so on and so forth.

I think once you get down to a measured value of about a milliohm or maybe hundreds of microhms you will find that this becomes a somewhat tedious undertaking, at which point you will have to decide whether to press onward in search of more measurement accuracy or declare good enough.  One of the things I like about the copper wire test fixture approach is that it's very easy to adjust the length of a wire (assuming you can measure and cut it accurately and assuming you were able to find the correct values for the theoretical wire), and then you can make the math easy when you do your hypothesis testing by shortening or lengthening the wire for more tests.  I think this offers some advantages over using a reference resistor as it's somewhat hard to know just how accurately that resistor was manufactured.  With just the copper wire it's easier to figure out what the value should be, plus unlike a reference resistor you can tweak the reference wire's resistance by cutting different lengths.

Once you get your baseline "DUT"-wire (or several wires of different lengths) established with your initial set of clips/leads, then you can move onto your next set of clips/leads and see what variations you get in the measurements, and if you have the desire you can compare lots of clips/leads.  Of course, you could also use a preferred set of clips/leads to then compare meters.  By this time you will probably need a way to control or at least account for changes in the ambient temperature.

This approach can be enlightening and fun as long as it maps reasonably well to your level of OCD.

[tooki]

So I ordered some probes from Ebay and AliExpress and now am waiting for arrivial (will take some time). I started thinking of a test setup/procedure to be able to tell which probes are bad/ok/good. Of course there is build quality (materials), ruggedness, ergonomics, etc. But what I am interested in for the time being is the electrical performance. I have a HP34401A which hasn't been calibrated in a long while and a DE-5000 LCR. I plan to use the kelvin clips with both of these instruments.
The question is: what is a standard procedure that can be used to assess the quality of the 4 wire kelvin connection on each one of these instruments? For example do people use a standard 1% resistor (which value?) or something better and measure it? What parameters do you setup in the instrument (nulling, NPLC)? How long do you people measure for? What post-processing do you apply to the measured data?

I think nobody answered because there’s no simple answer, and the question itself suggests you may not really understand the situation.

The entire purpose of Kelvin resistance measurement is to essentially eliminate from the measurement the effects of the test leads and clips. I don’t think any simple test with a resistor will be able to show any differences, because it shouldn’t by definition.

So in the end, the “build quality (materials), ruggedness, ergonomics, etc.” are what differentiates one from another.

Now, for capacitance and inductance measurements, it is probably different since they are frequency-dependent, but I’m honestly not sure to what extent. The test frequencies in a typical LCR meter (a few hundred KHz at the most) are simply too low for the cabling to be critical. Heck, the original Kelvin clips that came with my old Hioki LCD meter use proper 50 ohm BNC plugs, but with 75 ohm coax cable. It just doesn’t matter in this application.

[todorp]

The entire purpose of Kelvin resistance measurement is to essentially eliminate from the measurement the effects of the test leads and clips. I don’t think any simple test with a resistor will be able to show any differences, because it shouldn’t by definition.

I understand this, but this is theoretically. I immagine no leads and cables are perfect so there should be some way to say this pair is better than this one because it introduces less noise and gives a reading which is closest to the true value of a resistance with less variations... Of course "the true value" is also something abstract: that is why I am asking how people perform this kind of measurements and what they use as "the true value".

[GlennSprigg]

The entire purpose of Kelvin resistance measurement is to essentially eliminate from the measurement the effects of the test leads and clips. I don’t think any simple test with a resistor will be able to show any differences, because it shouldn’t by definition.

I understand this, but this is theoretically. I immagine no leads and cables are perfect so there should be some way to say this pair is better than this one because it introduces less noise and gives a reading which is closest to the true value of a resistance with less variations... Of course "the true value" is also something abstract: that is why I am asking how people perform this kind of measurements and what they use as "the true value".

For my old 2c worth, and just for interest!!.. haha...
Back when I was a Tech with Honeywell, field "temp Sensors" were wired using a '3-wire' system, that the electronic/digital instruments REQUIRED,
to counteract the cable resistance to the Field. Some sensors had values of about say 100-ohms, at room-temp. A 5 or 10 Ohm cable resistance, for say
a hundred Metre cable to the field sensor obviously introduced huge errors, for the sensors temp range. So, 2 of the wires were shorted together at the
sensors. The 3rd wire went to the other side of the sensor. The 'Instrument' would measure the resistance of the shorted Loop up & back, and then
subtract that resistance from the resistance of one of the shorted loop wires, returning on the 3rd wire, as that should be the same! That always cancelled
out any cable resistance, no matter how long the cable is!!, so it then only saw the actual sensor resistance for calculations...
But THAT was all done within the digital instrument/device, with nothing special out in the field.  Hope I didn't bore anyone!!   

[tooki]

The entire purpose of Kelvin resistance measurement is to essentially eliminate from the measurement the effects of the test leads and clips. I don’t think any simple test with a resistor will be able to show any differences, because it shouldn’t by definition.

I understand this, but this is theoretically. I immagine no leads and cables are perfect so there should be some way to say this pair is better than this one because it introduces less noise and gives a reading which is closest to the true value of a resistance with less variations... Of course "the true value" is also something abstract: that is why I am asking how people perform this kind of measurements and what they use as "the true value".

Again, talking only about resistance:

So… you want to measure something that, while measurable, doesn’t matter because it doesn’t influence the measurements you’ll make?

[todorp]

So… you want to measure something that, while measurable, doesn’t matter because it doesn’t influence the measurements you’ll make?

Not so - if your sense cables are made from nickel I suppose your measurement wouldn't be very good, would it?

Also why would people go to great lengths to do the following if it didn't matter at all?
https://www.eevblog.com/forum/testgear/4-wire-2-wire-resistance-tests-various-cheap-kelvin-clips-various-meters/msg3240888/#msg3240888
https://www.eevblog.com/forum/testgear/the-ultimate-kelvin-grabbers/
https://www.eevblog.com/forum/testgear/small-kelvin-clips-that-actually-grab/
https://www.eevblog.com/forum/testgear/franky_s-kelvin-clip-kit/?PHPSESSID=h41h3jdg9eqpu3afq59s2upad4

Remark, am no expert or anything. Just trying to understand...

[tooki]

So… you want to measure something that, while measurable, doesn’t matter because it doesn’t influence the measurements you’ll make?

Not so - if your sense cables are made from nickel I suppose your measurement wouldn't be very good, would it?

What makes you think your measurements wouldn’t be good if it used nickel wire? The resistance of the leads basically does not matter (within reason; the 34465A manual lists a maximum resistance per lead of 10% of the range on the lower ranges (so 10 or 100 ohms), 1k ohm for the higher ranges.) Whether the leads have 100milliohms or 1 ohm resistance literally does not matter.

You have to understand that the drive lines drive a controlled, known current through the DUT, meaning that the current is a known value regardless of what resistance it’s going across (DUT and leads). The sense lines, in turn, measure the voltage drop across the DUT only. The sense inputs (the V input of the meter) have a  very high input impedance: 1Mohm on cheap meters, 10Mohm on good handheld meters, and much, much higher on bench meters: “>10Gohm” on the 34465A. The error introduced by a few ohms in the leads literally does not matter: essentially no current flows, so there is essentially no voltage drop in the leads.

Also why would people go to great lengths to do the following if it didn't matter at all?

People do all sorts of things that aren’t necessary.

Most of those reviews are about the build quality anyway.

Remark, am no expert or anything. Just trying to understand...

Nor am I, and should an actual expert chime in and tell me I’m wrong, I’ll listen.

[bdunham7]

Nor am I, and should an actual expert chime in and tell me I’m wrong, I’ll listen.

I won't assume the 'expert' mantle, but due to the practical construction of 4W ohms circuits, specifically that the SOURCE and SENSE are not completely isolated, the resistance of the SOURCE leads does matter--so you're just a bit 'wrong'.  As you say, any error due to resistance of the SENSE leads should be infinitesimal.  The good news is that unless that resistance is unreasonably high, your error is a couple of digits off the right hand side of the scale, so nobody will know.  The maximum resistances allowed by the specifications of the bench meters I've seen are much higher than they need to be--there's no reason to exceed 1 ohm, IMO.  I suppose they are just basing that on the point at which the error will become apparent.  The big issues seem to me to be contact resistance/noise/thermal voltage and picking up magnetic fields in the lead loops.  So I'll rate your statement 99% correct.  

[todorp]

@tooki - right, I understand all of this. In my previous post I was talking about the SENSE leads. My next question would have been: why not use some of the crappiest and cheapest conducting materials for the SOURCE and SENSE leads if they do not matter?
Thanks to @bdunham7 this question is answerd.

[Shock]

The answer to your question is *everything* counts but it entirely depends on what you are measuring, how low/high the measurement is, the performance of your instrument and technique. It's such an involved subject it's easier to just go research by reading than explain it all.

If you look at various commercial offerings you will note some are guarded which gives them a higher bandwidth and reduces some of the problems associated with longer leads. Genuine cable (not cheap chinese) and quality clips/plating can get quite expensive. Considering it's outside your budget, avoid worrying about that until you need it. I would start by making or buying a set of $20 leads.

Aside from good clips, keep an eye out for chinese leads that only solder to one side. Cable flexibility is important to some as well.

[Electro Fan]

The answer to your question is *everything* counts but it entirely depends on what you are measuring, how low/high the measurement is, the performance of your instrument and technique. It's such an involved subject it's easier to just go research by reading than explain it all.

These last several posts and this post are why I think it makes sense to get yourself a piece of copper wire that you can reasonably characterize and quantify and use it as a benchmark test bed as you try various off the shelf and/or DIY 4 wire clips/leads.  You can see how the clips/leads perform relative to the theoretical spec, how they compare to other clips/leads, and even how they compare to themselves as you strive for repeatability - which will reveal some of the challenges (clip grip on the wire, precise clip locations on the wire, settling time, temperature fluctuations, test durations, and possibly other variables).

Once you get your benchmark test wire and your benchmark clips/leads established with the first meter you might find it interesting to then try another meter.  This can go on and on until you are pretty far into a metrology rabbit hole.  On the upside, there is a lot to be learned by just doing measurements, developing a hypothesis, and testing some more. 

Net, net:  The copper wire test bed costs close to nothing and it can be a fun opportunity to learn from both research and from doing.

[guymo]

I won't assume the 'expert' mantle, but due to the practical construction of 4W ohms circuits, specifically that the SOURCE and SENSE are not completely isolated, the resistance of the SOURCE leads does matter--so you're just a bit 'wrong'.

I am feeling stupid right now because I can't figure out what error this could cause or how it would arise. Would you (or someone) mind explaining a bit please?

[tooki]

Ditto. That doesn’t make sense to me. Eliminating the effect of resistance in the source leads is critical to the concept of kelvin resistance measurement.

[mawyatt]

Using a very simple model for the Kelvin resistance measurement where the DMM meter voltage reading impedance is assumed to be infinite, and the DMM current source output resistance is Rs. Using equal lead resistances from the current source of Rl, one arrives at a simple equation for the Resistance under test Rdut as:

Rdut ~ (Vm/Is)*(1 + 2Rl/Rs), where Vm is the measured voltage across Rdut and Is is the source current.

If Rs >>> 2Rl then Rdut = Vm/Is as expected.

The change is Rdut as dRdut is:

dRdut = dRl/(Rs/2 + Rl), where dRl is the change in source lead resistance.

Again if Rs >> Rl then dRdut is small approaching ~>0 as Rs ~> infinity.

Don't know the value of Rs but suspect it's quite high in quality DMMs, then the error introduced by the source lead resistance should be small.

However, all this "assumes" the DMM current source is ideal and constant except for the output resistance Rs. It's likely the current source value does vary some small amount with the variation in Rl which causes a variation in current source output voltage. 

So it seems the common sense approach as mentioned by many would be to keep the source lead resistance small but not go to extremes. 

An interesting experiment would involve a precision, stable, low value test resistance and the Kelvin clip source leads arranged to add a series resistance in each lead to observe the effects.

Best,

Edit: Just did a quick test with the old HP34401A  and small cheap Kelvin clips. Added a series R in each leg of the Source Lead of 1 and 10 ohms. DMM was Nulled for zero reading with clips shorted to device lead in first reading and not again.

Results:

R value        Kelvin Reading            1 ohm Rlead              10 ohm Rlead
1 ohm         0.9966                        0.9966                      0.9964
second        0.9962                        0.9965                      0.9962
1K              0.996596K                   0.996596K                0.996596K
1 ohm         0.9454                        0.9450                      0.9453
120 ohm     119.9989                     119.9988                  119.9989
2 ohm         2.0002                        2.0002                      2.0002

Added some reading from KS34465A

1 ohm        0.9972                         0.9972                      0.9972
2 ohm        2.0020                         2.0020                      2.0020
10 ohm      9.9469                         9.9469                      9.9469
120 ohm    119.9941                      119.9941                  119.9937
1K             0.996636K                    0.996636K                0.996636K
1 ohm*     997.13302m                  997.13513m              997.12309m

1 ohm* is using scaling y =mx +b and stat mean of 10 readings

[bdunham7]

I am feeling stupid right now because I can't figure out what error this could cause or how it would arise. Would you (or someone) mind explaining a bit please?

Ditto. That doesn’t make sense to me. Eliminating the effect of resistance in the source leads is critical to the concept of kelvin resistance measurement.

Didn't make sense to me at first for the same reason, and although what mawyatt says is true, it isn't what I ran into.  The base expectation is that the current source will provide a precise current up to a compliance voltage and that the SOURCE and SENSE are isolated, with the meter determining the reading by measuring the SENSE voltage and then doing some math and calibration corrections.  You might do a similar measurement with an independent power supply and millivolt meter.

I'm trying to make (as in produce) a type of 4W probe fixture and what I ran into on at least some meters I worked with is that the SOURCE and SENSE are not completely isolated nor independent.  If you short the SENSE leads with the SOURCE open, you do not get a zero or very low reading, as you might expect.  I have not fully characterized or disassembled this particular meter nor do I have a schematic, but I can tell you that it does not seem broken and works very, very well when used as intended.  The specifications call for the 4W SOURCE lead resistance to be "less than 10% of range" for the 10R, 100R and 1K ranges, so 1 ohm in practice.  The test current in the 10R range is 5mA and the current source OCV is over 5 volts, so I have to assume that this is not in any way a source impedance or compliance voltage issue.

I'm sorry I don't know more, but if I ever get my probe samples made I'll have another look in detail and report back.