Even More on Current Sources and a Kelvin (4-Wire) Milliohm Meter

Hi Frank!

I'm wondering if it is possible to use a dual-stage design, i.e. separate the current source from the final signal output stage, i.e. use an op-amp for the voltage amplification across the load, but perhaps with a lower-current source, since 100mA might make it more difficult, because the reference resistor will get warm and maybe drift. That op-amp you mention (in a min gain of 10 version) could also be used for the second stage.

Any of the current source ideas could be used for the first stage, but I think the LM334 would be attractive (but means your first stage would not be an op-amp, so depends if this is a hard limitation or not : ) because the power in the resistor is really small (since the internal reference inside it is 68mV), i.e. if a 6.8ohm resistor is used, power dissipated in the resistor is 0.68mW.  Then for the desired 1mohm to 10 ohm range, if the meter is 0-2V (as an example) then a gain of 20 is needed (i.e. compatible with the min. gain of 10 version). Or for an op-amp version, it could be the same as the circuit you have now of course (with the reference resistor value changed). These are just some ideas, maybe unnecessary if the resistor won't drift much (or if you find in the end that less than 100mA is fine too, for the particular multimeter you use).

Hi Shabaz,

An update... I read the datasheet thoroughly and set up the LM334 as a temperature compensated current source as described in the datasheet.  A IN4148 was substituted for the diode they used.  I came up with nonstandard resistor values so I put the closest values I had in.  The initial test was about 2% off the desired 10 mA current (not unexpected based on what was in the datasheet) so R1 was adjusted by adding another resistor in parallel until within 1%.  Voltage was varied over the range of interest and here are the results for the full length of test wire:

Things look good above 2.5V input (the planned instrument will have at least 3V) but notice there is a slight rise in the measured current at 5V.  This may be due to temperature from increased voltage which the datasheet warns of or just increase of temperature as it heats up in time in still air.  I left it on for an hour at 5V and from that point on it was stable although I am unable to make measurements with the resolution I would like.

I added another tap in my test wire at quarter length.  The results are:

Full Length:  0.092 Ohms

Half Length: 0.045 Ohms

Qtr Length:  0.022 Ohms

The accuracy appears to fall off as it nears the bottom of the multimeter range but all is as expected.  I will look into adding amplification next.

Hi Frank,

Good point! I'm not sure to be honest. I've not seen ones with a more precise definition. Usually when they state (say) 1V, that's only likely at a certain current anyway, since it changes depending on current (lower current results in lower Vf to a limit). I do have the diode they mention in the datasheet since I ordered that at the same time as the LM334, but I will also try 1N4148 so we can replicate.

Here's what I'm currently attempting (basically same circuit as yours, and including the BJT that was in the EDN article. Also the datasheet (at least the Linear Tech one - although I'm using a TI LM334, not LT) mentions a resistor and capacitor for stability, which I sketched, but I'll leave out for now until I've read a bit more.

I've quickly prototyped it on a small breadboard, and tried it with voltages from 3-11V, and it seems stable-ish, I see a difference in current of about 10uA as the voltage is varied from 3-11V, but even moving the distance between the BJT and LM334 makes an improvement : ) so maybe it is worth spreading out the circuit by a few holes in the breadboard could make a difference : ) I didn't run the test very long, it's very experimental : (  I need better parts to be honest, and then test indoors and outdoors, to see what change there is with the ambient temperature changing.

I'm using cheap 5% resistors (from a Velleman kit, so no detail like ppm/degC spec), and I used 12 ohm and 120 ohm (matched with a multimeter as best as I could manage, which wasn't all that well) instead of the values in the diagram below. Anyway even mW level they will get warm slightly, so better quality resistors are ideally needed. The current does creep up (by a few tens of uA) over the space of the few minutes that I tried, but I suspect that's due to the resistors warming up, because it did not seem to accelerate if I increased the voltage. So big (or several in parallel) and more stable resistors could help : )

EDIT: Just as an experiment I also tried 120 ohm and 12 kohm combination, for about 1.154mA output, and it's stable to within 3uA, for the range 3-11V, over the several minutes that I ran it. But I still have no idea if it is stable versus ambient temperature.

Hi Shabaz,

I have been wondering what kind of mischief the use of a breadboard might be causing.  I might solder something up and use sockets for the expensive components.  Will also try adding a transistor and redo the math on the resistors and probably change them.  I hooked up the MCP6N16 with calculated gain of 101 and it introduced about a 4% additional error but I know I am too close to the output rail.  I need to go back see what I can do to fix that.

Great idea. I'll try sourcing a few slightly better resistors meanwhile, although it may have to wait until there are enough things to order : ( since I already ordered some stuff yesterday.

By the way, if you've only so far read the TI datasheet, it could be worth checking the LT one too, I just noticed it says to locate the resistors close to the LM334, and also to avoid a socket on that particular part. It's all quite interesting how much care is needed for each component and location : )

I haven't looked at the LT datasheet.  Will do that...  When I ordered the LM334 I ordered a through hole and SMD version so I have an extra to play with.

Hi Frank,

I was reading last night that using a couple of legs of a BJT as a PN junction could be more consistent than using a diode (PDF doc here:   http://www.ti.com/lit/an/sboa277/sboa277.pdf  )

I'm going to try measuring the Vf at 5mA (since that is approx what goes through the diode in that circuit) of a few popular TO92 BJTs (I'm using five, two are 2N3904 (from different manufacturers) and three are BC547 from different sources too, and I've recorded the exact part numbers. I'm not too experienced with this, I've waterproofed them with epoxy glue (photo below shows them without the epoxy) and will put them in boiling water (and measure that with a temp probe too), i.e. just two datapoints for now, room temp and ~100 degrees C. I'll also compare with the diode they mention in that datasheet, that's on the left side of the array in the photo. It's a bit of a simple test : ( and different batches may be different : ( but for now I'll see what happens with just these components, and report back : )

Hi Frank,

I was reading last night that using a couple of legs of a BJT as a PN junction could be more consistent than using a diode (PDF doc here:   http://www.ti.com/lit/an/sboa277/sboa277.pdf  )

I'm going to try measuring the Vf at 5mA (since that is approx what goes through the diode in that circuit) of a few popular TO92 BJTs (I'm using five, two are 2N3904 (from different manufacturers) and three are BC547 from different sources too, and I've recorded the exact part numbers. I'm not too experienced with this, I've waterproofed them with epoxy glue (photo below shows them without the epoxy) and will put them in boiling water (and measure that with a temp probe too), i.e. just two datapoints for now, room temp and ~100 degrees C. I'll also compare with the diode they mention in that datasheet, that's on the left side of the array in the photo. It's a bit of a simple test : ( and different batches may be different : ( but for now I'll see what happens with just these components, and report back : )

Hi Shabaz,

Great idea to use a BJT.  TI writes some really good applications reports.  I like your test setup and eagerly await the results.  Really clever.  It would be fun to get intermediate points closer to room temperature. If the pot of water is large will it hold temperature as it slowly cools long enough to approximate steady state and get some readings?

Hi Frank!

Unfortunately I was in the kitchen, doing the experiment : ) I didn't see your message in time regarding intermediate readings : ( and I've dismantled it from the kitchen now.

I did want to do that, but in the end I didn't because the pot I'd used wasn't so large and it was taking me 60-90 seconds to do the readings for the 6 components (5 BJT and the diode) since I was swapping wires manually and waiting for things to settle - so wasn't sure how fixed the temperature would remain. But I may do it again since it's a good idea. But I did do the measurements at 6 different currents, so we have some numbers to play with anyway : )

Anyway, here are the raw results : ) I've not examined them yet : )

There was quite a lot of bubbles (I used UK's southern water company's finest tap-water: ) it's not very great : ) Full of minerals and "stuff" : )

Anyway, the water temperature did fluctuate throughout the experiment, but only by about 0.2 degrees.. I measured things twice, so we can also see if the fluctuation in the measurement is significant too.

I've got erroneous decimal places in the first few results, so the results are of course 0.6734V, 0.6916V, 0.7025V etc.. for the first column, and so on. So, first set of results are at room temperature at 1-6mA, and then the next two sets of results are at 99.82-100.0 degrees C, at 1-6mA again.

I'm off to find some actual food in the kitchen : )

The temperature coefficient results : )

These were the components tested:

And the results (i.e. change in Vf, in V/degC):

The TI document (http://www.ti.com/lit/an/sboa277/sboa277.pdf ) had this chart for a typical transistor:

The results are extremely close to the TI doc. It doesn't matter which BJT or diode is used, it will just affect the ratios of the resistors in the circuit anyway. But at least now we can more precisely set the ratio. The LM334 datasheet had suggested the IN457 had a tempco of -2.5mV, but I see -1.6mV at 5mA as you can see in the results, so the ratio wouldn't be the nice 10:1 that it is in the datasheet example.

Just to extract what else I could find from the results.. it's interesting, in terms of forward voltage, both 2N3904 were near-identical, despite one coming from On Semi and the other from Nat Semi.

However, the three BC547 variants had a clear separation between those from On Semi (which had very close behavior) and the one from Fairchild.

In summary, for 2N3904, the manufacturer didn't seem to make a difference, whereas for the BC547, the results were similar only if the variants were from the same manufacturer. This is a very weak summary since the sample quantity is so low, but still interesting!

Nice... That is really good stuff, and as you point out the results are quite close to the TI doc for the BJTs tested. It seems like it would be good to provide for ratio adjustment by a trimmer pot.  When I was doing my tests above I put another resistor in parallel to bring it within 1%.