LCR impedance measurement repeatability (Agilent 4285A)

[mtwieg]

I made a setup to characterize litz wire performance, and am having some issues with measurement repeatability I can't explain. I'm using an Agilent 4285A (uncalibrated, but in good condition).

The DUT for now is a simple air core solenoid wound on an acrylic former. This DUT is just something I threw together to refine the measurement setup/procedure to my satisfaction. My intended frequency range is 100kHz-10MHz, measurement mode is Ls-Rs.

To make the connections from to the DUT as robust as possible, I use the 16047A test fixture instead of the kelvin clips. I also created little adapters boards which fit nicely in the "jaws" of the test fixture (see attached photo). These adapters are made of double sided copper clad, with the bottom/top layers (force/sense terminals) connecting at the far edge, which is also where I solder the litz wire for the DUT.

This setup seems much better than sticking the litz wire directly into the jaws of the 16047A, but I'm still seeing odd variability on the Rs result. For example, I plug in the DUT+adapter, and at f=1MHz I measure Ls=15.98uH +/- 0.05uH, Rs = 0.251 +/- 0.015ohm. But if I then wiggle the adapter a bit in the jaws, the Rs measurement will settle at a very different value, like Rs = 0.331ohm, but measurement noise is still +/- 0.015ohm.

I'm fairly certain the only thing moving is the adapter itself (the DUT coil is mounted rigidly and there's nothing nearby for it to couple to). The contact resistance on the four terminals is probably varying quite a bit, but my understanding is that so long as all four terminals of the instrument make "decent" contact with the adapter (i.e. impedance low enough that it doesn't cause clipping in the current drivers, or cause phase shift in the sense signals), it shouldn't affect the final measurement.

Another explanation is that strands of the litz wire are intermittently breaking as I wiggle the adapter. But if this were the case I would also see it affect DC resistance, and I've verified this isn't the case (used a 4263B for that).

[KungFuJosh]

My guess is that your adapter is at fault, but you should also be using a guard plate beneath the DUT.

If you can use two solid pieces of copper instead of what I'm assuming is copper tape, that might improve the connection quality with the fixture.

If you have good kelvin clips, I would also consider properly adjusting them, and seeing how it looks with that when using a guard plate.

[TimFox]

How did you terminate the two ends of your Litz wire?

[mtwieg]

My guess is that your adapter is at fault

That's my guess too, but not sure why or how to improve it. Except just giving everything a thorough cleaning with IPA and polishing.

but you should also be using a guard plate beneath the DUT.

I had come across this in keysight's impedance measurement handbook. My understanding is this helps reduce the effects of stray capacitance, but if stray capacitance were an issue I would expect that to affect the Ls measurement more than Rs... but a large ground plate will likely interact strongly with the coil's magnetic field. Still worth trying, just to see.

If you can use two solid pieces of copper instead of what I'm assuming is copper tape, that might improve the connection quality with the fixture.

The adapter is made of double sided copper clad. The copper clad is what makes contact with the jaws, and I just use a sliver of copper tape at the other end of the adapter to connect the P and C sides together (also soldered).

I wonder if the large contact area of the adapter is backfiring by reducing the contact pressure, making it less likely to break up oxides on the surface. Going to try cleaning the surfaces, and maybe narrowing the legs of the adapter to reduce the contact area. Wish I could put gold plating on the adapter legs. Maybe should order an actual PCB with ENIG plating...

If you have good kelvin clips, I would also consider properly adjusting them, and seeing how it looks with that when using a guard plate.

I do have a set of 16089B kelvin clip leads, but typically they don't work very well with a DUT like this (unless the clips are mounted somehow). Clipping them to the adapter plate might help though.

I'm not aware of any adjustments I can make to these, aside from maybe replacing the springs.

[mtwieg]

How did you terminate the two ends of your Litz wire?

Pretty carefully. I tinned them with a solder pot and a bit of flux.

The measured DC resistance per length is just about 6% higher than what I expected based on the wire's "equivalent" AWG specification. To validate my tinning process, I cut off the tinned part, re-tinned, and re-measured Rdc several times on the same length of wire, and I get at most 1-2% variation in resistance. And no observable change when bending the wire. So I think that's good.

For attaching the wire to the adapter, I just used the minimal amount of solder and heat. Don't want to disturb the arrangement of the strands (I intend to play more with this later, but not until I get my measurement setup figured out).

[KungFuJosh]

If you can use two solid pieces of copper instead of what I'm assuming is copper tape, that might improve the connection quality with the fixture.

The adapter is made of double sided copper clad. The copper clad is what makes contact with the jaws, and I just use a sliver of copper tape at the other end of the adapter to connect the P and C sides together (also soldered).

I wonder if the large contact area of the adapter is backfiring by reducing the contact pressure, making it less likely to break up oxides on the surface. Going to try cleaning the surfaces, and maybe narrowing the legs of the adapter to reduce the contact area. Wish I could put gold plating on the adapter legs. Maybe should order an actual PCB with ENIG plating...

I've found copper tape unreliable with this type of setup. This is probably partly due to the lack of conductive mass. The short bars are thick for a reason.

I think the PCB with ENIG is a cool idea and worth trying. I'm tempted to make one too just for entertainment value. Might come in handy in the future.

If you have good kelvin clips, I would also consider properly adjusting them, and seeing how it looks with that when using a guard plate.

I do have a set of 16089B kelvin clip leads, but typically they don't work very well with a DUT like this (unless the clips are mounted somehow). Clipping them to the adapter plate might help though.

I'm not aware of any adjustments I can make to these, aside from maybe replacing the springs.

I meant cable/open/short adjustments/corrections in the LCR. I assume it at least has open/short correction. Sometimes it's called adjustments, or correction, or calibration.

Thanks,
Josh

[mawyatt]

Try and use a physically large resistor, with a very high resistance to not effect the measurement. Carefully wrap the Litz wire around the resistor leads near the body and solder, then use the Standard Leaded LCR Meter Fixture for measurement.

Best,

[mtwieg]

I've found copper tape unreliable with this type of setup. This is probably partly due to the lack of conductive mass. The short bars are thick for a reason.

I doubt the conductor thickness is an issue at this frequency. The surface quality is probably more relevant...

I meant cable/open/short adjustments/corrections in the LCR. I assume it at least has open/short correction. Sometimes it's called adjustments, or correction, or calibration.

Oh yeah I'm doing open/short correction. Maybe I should consider adding a load to the correction as well.

Reading more into Keysights impedance measurement handbook, it suggests the impedance of the load used for correction should be close to the DUT. But does that mean just the magnitude of the impedance? If my DUT is a coil, is it better to use a resistor or capacitor as the correction load? I happen to have lots of nice porcelain capacitors I could use, but not any fancy non-inductive resistors...

On the other hand, the handbook suggests that corrections should only be done if there's substantial unknown hardware in between the LCR meter itself and the DUT (like switches, additional cables, bias tees, etc). So maybe in my case where I'm connecting the DUT directly to the 16047A, maybe no correction would be better than a bad correction (assuming the LCR meter and 16047A are still in spec, anyways). Absolute accuracy might suffer, but at this point repeatability is the bigger problem.

Try and use a physically large resistor, with a very high resistance to not effect the measurement. Carefully wrap the Litz wire around the resistor leads near the body and solder, then use the Standard Leaded LCR Meter Fixture for measurement.

Yeah giving the litz wires a better fit for the 16047A should help. I'm going to try and find some header pins with gold plating and solder that to the litz.

[mawyatt]

Yeah giving the litz wires a better fit for the 16047A should help. I'm going to try and find some header pins with gold plating and solder that to the litz.

Think the header pins might be too short to fully engage with the fixture, here the larger physical size resistor will have better results as that's the type of component these fixtures are designed to interface with. Having a Gold plated, or Tin/solder plated lead shouldn't matter as it not going to be inserted often and contact oxidation isn't an issue.

Best

[mtwieg]

Got a chance to work on this more over the weekend.

Good news is that after thoroughly cleaning all the contacts of the 16047A with IPA, my initial problem with Rs repeatability seems to have been resolved. Found that results were consistent between using my adapter and inserting the tinned litz directly in the jaws.

Bad news is I noticed another confounding factor which is even more severe. Basically I noticed that if I put my hand on the coil, the measured Rs decreased (depending on the distance and area to my hand). At 1MHz, I could make Rs drop from 0.25ohm to 0.12ohm. I then changed the frequency to 3MHz. Rs was normally 0.88ohms, but with my hand wrapped around the outside of the coil, Rs would actually go negative, as much as -8.5ohm. This doesn't seem to be related to any stress/force on the DUT due to my hand (putting a 1cm thick layer of foam between my hand and the DUT greatly reduces the effect, while still putting similar forces on the DUT). Just some strange result of fringe fields.

I took the time to use a VNA with pickup loops to accurately measure Q at 1MHz and 3MHz. Found Q=554 at 1MHz (Rs=0.182) and 637 at 3MHz (Rs=0.473). So even under best conditions the 4285A is off by a factor of two at 3MHz.

I took a step back and calculated the specified accuracy for the 4285A for my frequency/impedance. For 1MHz and Q=554, Rs error is +/-0.128ohms, Q error is +1311/-229. For 3MHz and Q=637, Rs error is +/-0.787ohms, Q error is +1600/-398 (actually more like +infinity/-398). I'm fairly certain I did the math right, but those numbers are terrible. I'm assuming one can get much better accuracy with very careful load corrections, but I don't think I want to go down that rabbit hole. For now I'll rely on the LCR meter just for apples-to-apples comparisons, but the VNA will be the final say when it comes to Q and Rs.

[mawyatt]

Your hand is creating a complex and variable impedance with the DUT coil. For repeatable and realistic results you need a constant environment and setup.

If you study how Lab Bench LCR meters operate, with high "Q" components the measurement phase approaches +- 90 degrees which is highly sensitive region for computing the various DUT parameters and prone to error and complete nonsense. The measured magnitude of "Z" is usually good but the distilled values for the equivalent component (R) for loss is highly sensitive.

If one thinks about what's happening under these conditions, anything such as a slight change in measurement fixture length (moving a cable or wire), or DUT contact point (effects DUT length), or environment (hands and other stuff) can slightly change the measurement angle, and around 90 degrees for high "Q" DUT this significantly influences the loss component computed from the complex impedance Z.  We often "see" results of negative R with very high "Q" components with our IM3536.

Best