PCB for a Kelvin (4-Wire) Milliohm Meter

I started building this tonight : )

I've soldered all the resistors and pots, and then trimmed to the desired resistance.

Hopefully tomorrow I'll get the rest all soldered up, and try powering it up before putting it in the case!

I was just discussing the project with my fellow geek friend this evening and told him I hoped you would have a chance to look at it soon.  Let me know your thoughts and upgrades and I will incorporate when I respin the board.

Thanks for all your comments and help Shabaz!  They are more than cosmetic.  I plan to make another pass on the board and will incorporate your thoughts.

RE Current Error:  Too long ago to remember but my comment says I had to reduce the resistance which would mean I was getting low current I think.

I calculated new resistor values by running my spreadsheet at higher current as you suggest above

Hi Frank,

Problem solved (I think) : )

Somewhere along the line the tolerance of the LM334 was missed, we probably both assumed it was absorbed into the calculations once we'd created the spreadsheets, or that it would be trimmed out by the existing trimmer circuit, and didn't examine the datasheet again. But, although it is stable, it might not be at 10mA. Although it can be adjusted out, really two resistances need to be trimmed for 10mA, and two for 1mA, otherwise the temperature compensation will get affected : (

Here's an alternative suggestion, and it seems to work.. basically, instead of trimming current to 10mA, the gain of the amplifier could be trimmed instead. I think we don't need the current source trimmers at all, since it seems to be stable with no adjustment. So, no more trimmers are needed, but they are instead used for amplifier gain adjustment (plus this is better for the future too, since if it is ever modified for ADC, then zero trimmers are needed since the measured value can be multiplied by the microcontroller).

Since there are two ranges, that means that amplifier trimming also needs to be switched along with the range. But that's ok, it is possible to do it with a normal DPDT switch with one pole selecting the current source, and the other pole switching in the second trimmer. It means the second current source should be changed from 1mA nominal, to (say) 1.1mA, because then it's easier to have such a trimming mechanism without more switch poles.

The mod below shows the idea.. the 20k trimmer is used to adjust the reading for the 10mA current source. Then, the switch (one half of the DPDT switch) is closed, and the 500k trimmer is used to adjust for the 1.1mA current source. The values were chosen so that a +-3% gain adjustment can be made via the trimmers, for both ranges. The 10mA gain adjustment has to be trimmed first with this topology, since the 20k trimmer is never switched out. The 500k trimmer will work regardless if the 20k trimmer is set at any extreme, if the current source is designed for 1.1mA +-3%.

I don't have a 500k trimmer, so I just tested the 20k trimmer, and the measurements are spot on for 116mohm, and 14mohm (these are the only two values I tested so far) with sub-milliohm error. Since both of my current sources have approximately the same proportion error, I actually don't need the 500k trimmer switched in if I keep the second current source at 1mA instead of 1.1mA, but that can't be assumed for all builds of course!

This is what my build looks like currently, i.e. just the 20k trimmer mod, and no switching since the error adjustment happens to be the same across both ranges:

Maybe for the rev 2 PCB, as well as these new amplifier trimmers, the existing trimmer locations are kept, but are set to do-not-fit for now, since all seems stable without them.

Here are the values for the 10mA and 1.1mA current sources:

10mA: use -1.85mV/degC and Vf=0.7199

top resistor network to equal 17.7931ohm, which is 33.2 ohm paralleled with 38.3 ohm

lower resistor network to equal 127.2167 ohms which is 160 ohm in parallel with 620 ohm

1.1mA: use -2.01mV/degC and Vf=0.660V

top resistor network to equal 145.8354ohm which is three resistors, 150 ohm, 5.6k and 82k all in parallel

lower resistor network to equal 1145.483 ohms which is 1.33k in paralled with 8.2k

With the current schematic, R7 and R22 and trimmers R8 and R23 would become DNP.

I'm going to box it all up now, see if it still functions : )

Hi Shabaz,

That is very clever!  I am more committed than ever to a new revision.  It needs to be something that even John Wiltrout would be proud to put on his bench :-)

I suspect you are correct on the issue and on looking back at the datasheet I am sure I did not appreciate it at the time and was thinking it could be corrected with the trim pots (in fact it was very close on the protoboard and easily trimmed). Since we are rethinking, I have a couple of thoughts...

• Will this create a tempco issue at the instrument amp?  I need to work through this.
• Is tempco really an issue?  If it is calibrated at say 25 C will the temperature change by more than say 10 C which should be OK?
• Can the scale switching problem be easily solved - i.e. having to scale by factor of 100 in ohm range - this point was CORREcTED for earlier brain freeze
• Can the comparator switch a MOSFET for automatic range setting?

I see there is still room for a larger board in your enclosure (which I would like to use).  The next version could have:

• proper mounting holes that match the enclosure.
• Would ~6 V battery power / holder fit in the case?
• What about including a 1 ohm calibration resistor into the design with a jumper

Mostly just thinking out loud...

Frank

Hi Frank,

I had a chance to try it out a bit more, and this time tried four different resistances from 8mohm to 1.74 ohm, and in each case, the measurement is spot-on, under 1mohm error (I measured the resistances in 4-wire mode on the calibrated Keithley meter, to compare). The trimmer resistor in the amp gain section shouldn't cause much temperature effect I think, because it is only a small adjustment compared to the fixed resistors which set the bulk of the resistance for the gain setting. I've not tested for very long periods of time yet, nor at different temperatures, but at room temperature the values are stable within seconds. The tempco stuff was more of an issue for the current source, because 10mA is enough to self-heat devices, whereas the amp circuit is running a lot cooler so is only affected by ambient changes. But as mentioned the values are correct in the brief measurements so far, and I've not seen any drift. Tomorrow I'll try to record a short video showing the current behavior of it.

Incidentally, there is some instability : ( Very occasionally when connecting a resistance, the output will go to almost zero (0.0008V) and I don't know what causes it. Disconnecting and reconnecting the resistor under test seems to resolve it. But it's not occurred often enough to get to the bottom of it yet : ( The actual resistance being measured doesn't seem to be the cause, because I've seen it occur when measuring different resistances. I'm wondering if the amplifier circuit is temporarily going funny, e.g. maybe something is on the threshold, or maybe PCB trace layout around it, or the very long test leads (they are about 1 metre long). It could be related to the length of wires on the trimmer bodge I did on the PCB, although I did try to keep them short.

Regarding scale switching, there could be a setting on some panel meters to move the decimal place. Another option could be to have an ADC+microcontroller+LCD screen, but then it is a larger maybe revision 3 project.

The comparator can in theory do auto-ranging, there was an untested schematic in one of the blog post comments. That will need more parts, so might not fit a small PCB any more though. There is a bit of room in the case, but not enough for a PP3 or AAA's.. maybe a thin LiPo could fit, but that would be tight too. But a battery is a great idea, also an option could be having a DC connector to allow switching to a battery if no external power is applied.

The 1 ohm resistor is a neat idea too, although could need more PCB space, since 4 jumpers are needed to add it to the 4 Kelvin connections.

I'm excited to see the next revision of this project : )

Hi Frank,

I had a chance to try it out a bit more, and this time tried four different resistances from 8mohm to 1.74 ohm, and in each case, the measurement is spot-on, under 1mohm error (I measured the resistances in 4-wire mode on the calibrated Keithley meter, to compare). The trimmer resistor in the amp gain section shouldn't cause much temperature effect I think, because it is only a small adjustment compared to the fixed resistors which set the bulk of the resistance for the gain setting. I've not tested for very long periods of time yet, nor at different temperatures, but at room temperature the values are stable within seconds. The tempco stuff was more of an issue for the current source, because 10mA is enough to self-heat devices, whereas the amp circuit is running a lot cooler so is only affected by ambient changes. But as mentioned the values are correct in the brief measurements so far, and I've not seen any drift. Tomorrow I'll try to record a short video showing the current behavior of it.

Incidentally, there is some instability : ( Very occasionally when connecting a resistance, the output will go to almost zero (0.0008V) and I don't know what causes it. Disconnecting and reconnecting the resistor under test seems to resolve it. But it's not occurred often enough to get to the bottom of it yet : ( The actual resistance being measured doesn't seem to be the cause, because I've seen it occur when measuring different resistances. I'm wondering if the amplifier circuit is temporarily going funny, e.g. maybe something is on the threshold, or maybe PCB trace layout around it, or the very long test leads (they are about 1 metre long). It could be related to the length of wires on the trimmer bodge I did on the PCB, although I did try to keep them short.

Regarding scale switching, there could be a setting on some panel meters to move the decimal place. Another option could be to have an ADC+microcontroller+LCD screen, but then it is a larger maybe revision 3 project.

The comparator can in theory do auto-ranging, there was an untested schematic in one of the blog post comments. That will need more parts, so might not fit a small PCB any more though. There is a bit of room in the case, but not enough for a PP3 or AAA's.. maybe a thin LiPo could fit, but that would be tight too. But a battery is a great idea, also an option could be having a DC connector to allow switching to a battery if no external power is applied.

The 1 ohm resistor is a neat idea too, although could need more PCB space, since 4 jumpers are needed to add it to the 4 Kelvin connections.

I'm excited to see the next revision of this project : )

Hi Shabaz and Happy New Year,

Thanks for your thoughts on tempco.

I have been playing with it a bit more too.  I have not seen instability at higher resistances - I need to quantify, but say above maybe 10 mohm.    Along the lines of your thoughts on the amplifier circuit I attributed this to not having a negative rail and being so close to ground.  This could be completely wrong however as I am only guessing.  Can you try resistances in the 1 ohm range and then maybe lower to see if you see instability there?  I questioned the accuracy below 3 mohm so it is good that you have been able to get good results down to 1.74 mohm and can verify with the Keithley.

On the scale switching I had a thought around maybe it would be easier with the new amplifier topology you sketched but I need to think through it and draw it out.  It is just something I thought of on my daily walk and then questioned whether it was valid or not.  Same thought occurred with auto-ranging - need to think through it.  It doesn't work.

On the Kelvin resistor see for example the Ohmite FC4T series that implements 4-wire with four pads to the resistor on the PCB.  I also saw an article somewhere that describes doing it with a regular resistor but running separate current and voltage sensing traces to the two pads.

I need to finish off the LoRa project so I can get back to this :-).  I will start a new post at that time with an update on where things are since this one is getting hard to follow.

Frank