Dielectric Constant Measurement Practices

If you care about the loss part of the measurement, you should consider taking your measurements at higher frequencies. For example, at 20 kHz your 80 pF has a reactance of about 100 kΩ. If you want to measure a Q of say greater than 100 (or a D of less than 0.01) at 20 kHz, that's equivalent to a parallel loss resistance of greater than 10 Meg-ohm, which is near the instrument's limit of measurement. Its errors, which are due to slightly inaccurate phase measurements, can easily be positive or negative.

1 MHz is an industry-standard measurement frequency. But many instruments go to higher frequencies, such as HP's 4192A to 13 MHz and their 4191A all the way to 1 GHz (ditto for the newer versions of these instruments).

Thank you for the reply! Yes, I knew that dielectrics have a higher ESR at higher frequencies, so instead of measuring at varying frequency, I knew that the highest measurement frequency available was likely the only wise option (200kHz). I forgot to mention that I was measuring at 200kHz. Water for example is a pretty good dielectric but the D will go up at higher frequencies because the molecules can't change orientation quick enough. So, testing at a higher frequency is better for trying to see differences in dielectric properties of materials.

I wish we had a better LCR meter but I think this will be generally good enough. The maximum resistance that it reads is up to 999 MOhm but as I understand, the usable range isn't likely that high. But with the same setup, the problems would be the same with other measurement devices. How can I change my setup to be able to cut out these errors so that the D and resistance will no longer read negative? Will the only option be to measure with a smaller plate separation or with a larger error so that the measured values are larger (and thus error is a smaller factor)?

Interesting discussion, but expect a few protests when you say "Water...is a pretty good dielectric...". Most people make the equivalence between dielectric and insulator, which is not always true, since not all materials will polarize and store a charge under the influence of an external electric field.

It's always best to put the words "pure" or "highly distilled" in front of "water" when used in this fashion, otherwise you'll get questions like "...how come people get electrocuted in the bathtub...". The water everyone is familiar with has minerals and other impurities that turn it into a decent conductor (and a poor dielectric).

Yes! But the dielectric constant of water is something like 80.6 so it's much higher than non polar materials. It was just an example :) Water was the only thing that I could think of that was really common and had a high dielectric constant.

http://www.csgnetwork.com/dieconstantstable.html

But this goes back to the original question of how to reduce the measurement errors so I don't get those unrealistic negative values.

Please re-read my first post. Your machine determines the (very high) parallel loss resistance by making a phase measurement of your sample, compared to an intrinsic 90° phase shift for a pure capacitance. Small changes in the measured phase determine the calculated loss resistance. If a phase of exactly 90° is measured, the instrument calculates an infinite resistance. A slight positive or negative phase wrt 90° corresponds to a high positive or negative resistance.

You'd probably be happier looking at the absolute value of resistance than looking at the measured phase of the loss impedance. A careful calibration of your instrument might shift the measured phase slightly to a more accurate value, but really, the dielectric Q is very very high, isn't that enough for you to know?

Maybe you want to think of series esr (that's Xc / Q instead of Xc * Q), but really, the instrument has no way of knowing the loss mechanism. Either way it's trying to determine a value based on a very very small measured phase shift.

You're absolutely right, and after re-reading your first post I understand it would be better to measure with a higher frequency but unfortunately, I don't have anything more sophisticated available to measure with. I might be able to borrow a spectrum analyzer though. Thank you for the explanation. It makes more sense now.

The goal was to actually qualitatively compare water content between samples of something. The dissipation factor and capacitance should go up and the resistance should go down, but I never got around to testing more samples because of the negative numbers I was getting. I initially surmised that there was something wrong with my setup. The first sample was "dry". Perhaps though, the capacitance readout on the meter is still accurate, despite the erroneous D and R so perhaps if I continue making measurements and acknowledging the errors, there might still be useful data?

Very good points! Thank you. I didn't think about the contact issue. I read some research papers on dielectric measurements to estimate water content in grains or othr things like sawdust, pellets etc and they seemed to make that work quite well with no contact, albeit with very large plates (30cm^2). Conceptually, I'm trying to do something similar, but smaller.

I'm using what seems to be a quality, small diameter coaxial cable that isn't melting in the oven haha. Just moving the cables doesn't seem to change the device readout, but touching the sample holder does.

And after you asked what the air measurement was, I feel awfully stupid for not testing that! Seems obvious now you say it :)

I can't see your setup, but your lead wires may also contribute to the problem. In my previous life job I tested the starting of fluorescent lamps with various power sources. That sounds easy, but I finally determined that my setup was very important. And the test object is not the only path of conduction at those frequencies. You may have to experiment a little bit to eliminate undesired circuit paths. If you want to ask me more off line, click on my name and in the upper right part of your screen, you can e-mail me through this forum.

Thank you. I have the coaxial cable coming from the "sensor plates" coming out of the oven going to the meter. The meter has shielded clamps that I twisted around each other to reduce induced noise. The ground clamp goes to the shield of the coaxial cable of course. It's probably OK, but it would certainly be improved if I had the right type of cable to directly go to the meter so I don't have exposed clamps in the middle. I was hoping this would be a simple and easy test but it turned out not to be :)

Have you considered the possibility that temperature gradients across different metals in your test apparatus and leads might be part of the cause?

Good question... I will have to test that somehow. Maybe I need to make sure that everything is made out of copper for example, because perhaps copper/steel would act like a thermocouple. Since my first sample was "dry", heat shouldn't affect it significantly but perhaps my materials are. Very good point!

The fixture will have a lot to do with it, Are you gaurding and grounding the test fixture?

I believe the stainless steel lab oven is well grounded, so it should be shielded pretty well. Whether it's the same ground as the meter however, I did not check yet! :)

Try replacing your test cell with different capacitors of known value and see what your results are. When things don't look right, simplify!

20 to 80 pf is a very small capacitance. I am pretty sure that your LCR Meter does not have the resolution needed to measure both C and D at those small C's.

I recommend a REAL capacitance bridge, such as an old GR 1620. That was a great precision bridge, but it was expensive. An older type 716-C would work as well, and was much cheaper. Both should be available on the used market.

There is no such thing as negative D for a passive capacitor.

Also, resonance with your leads won't be happening at 20 KHz.

Rixter, good idea! Will do that monday first chance.

WoodwardDL, maybe you're right that it's not made for this measurement range. The owners manual gave a way to calculate the accuracy knowing the frequency and the impedance at that frequency. Supposedly the accuracy is 0.6% at 200kHz and up to around 150 MOhm (which seemed to be as high as this test went). I was told this should be a nice meter but then again, this is way out of my area of expertise :)

Good Luck. I kind of tend to agree with WoodwardDL, that your measurement errors are overwhelming what you're trying to measure. That should show up if you start with a large enough capacitance and work your way down.

My only other suggestion is to make your dielectric as thin as possible and as large an area a possible. A source of error is going to be non-uniformity in the dielectric thickness, especially as it is made thinner.

Well I did some tests as suggested and soldered a 47 pF capacitor on an identical coaxial cable. The meter does exactly the same thing as my test setup and reads negative resistance. We concluded it is because the phase measurement in the meter seems to have an offset when measured at 200kHz. Another division has the identical meter and it gave practically identical measurements.

I do have a question here: The meter has a setting for displaying the data as if it were a parallel or series capacitance-resistance circuit. My thought was it would be parallel, because it's a capacitor with a resistance in between the plates. Also, the data seems to correlate better with what I imagined (the real component of the impedance is very high). Which is the appropriate way of displaying the information. I'm aware that changing this setting doesn't actually change the type of measurement, but only the way it calculates it.

One thing that I found interesting was that if I put a 100 Ohm resistor in series, the meter always read positive numbers. It seems unscientific, but it at least makes the readout correct. In series measurement mode the extra 100 Ohm resistor in series took the resistance from about -40 Ohms to about +60 Ohms. This indicates an offset error, possibly from the phase measurement.

The conclusion is that this meter isn't very good, despite the price and german manufacturing. The device is, if anything, consistent and the capacitance appears to be quite accurate (despite the negative impedance readouts).

Are you using the original leads for the meter? If you switched out leads, perhaps the meter was intended to work with leads that have 40 Ohms more resistance than the ones you are using...

.

40 Ohms does seem like it would be excessive resistance for leads, but I thought I'd suggest the possibility.

The SHORT and OPEN calibrations are supposed to compensate for any offsets. Did you do them?

Take a look at this Kelvin accessory on PDF page 23: http://www.home.agilent.com/agilent/redirector.jspx?action=obs&nid=-34051.0.00&lc=eng&cc=US&ckey=1000002219-1%3Aepsg%3Aman&pubno=16089-90020<ype=&ctype=AGILENT_EDITORIAL

Note the cables are labeled Hcur Hpot Lcur and Lpot

For the OPEN test, Hcur & Hpot are connected together and Lcur and Lpot are connected together AT THE FIXTURE or at the end of the cables.

For the SHORT TEST everything is connected together usually to a nice big bar.

The Kelvin clips of this adapter have connections to each side of the clip, Note the two wires to each clip. Thus in the open test the two wires going to each clip are connected together when the clip is closed.

Your HM8118 has a similar adapter.

On the PDF page 38 of your manual: http://www.hameg.com/manuals.0.html?&no_cache=1&tx_hmdownloads_pi1[mode]=download&tx_hmdownloads_pi1[uid]=959

It tries to tell you what to do, The short and open test have to be done for all frequencies used (the 2 minute version) or SGL for a single frequency.

Open isn't explained well.

The instructions are written for the probes and not a fixture.

Thank you, I appreciate your reply. I've been doing the short and open calibrations at the chosen test frequency. I didn't check to see what the before and after measurements were.

We have the standard four coaxial cable test leads that came with the device and they are configured like in your link. Each side of the clamps have the conductor from one cable each. I assume that is to reduce measurement errors.

Beyond the clamps, I just have a short length of coaxial cable that is soldered directly to my measurement device. I'm going to try to make my own four coax cable that goes directly to what I'm measuring. I found some BNC video cables that appear to have the same connector.

How are you accounting for the capacitance introduced by the coax?

Strike my question, I believe the answer is in the previous two posts....open and short calibrations.

If you tell your instrument to display Z and Θ, the results may be more intuitive for your situation. Presumably your 47pF capacitor is a NP0 / C0G dielectric type. These have very low dielectric absorption (which causes capacitance to change with frequency). You could measure your instrument's Θ error for various frequencies and capacitances, etc., seeing how it deviates from 90º, and develop corrections to add to your water measurements. Calculate your capacitance from Z.

As for using BNC video cables, give it a try, but be aware that those are 75-ohm cables, whereas your instrument likely uses 50-ohm impedances. At least my HP bridges use 50 ohms.

That's good advice. When measuring Z and Θ it reads -90.778 degrees. So, there lies the problem :) It's reading beyond 90 degrees for some reason. I wonder what changes in the system when I heat up the test material to make it read differently. The capacitor alone makes it read wrong all the time. Both meters were identical.

The capacitor was one of those beige ceramic capacitors. That is very sound logic, to measure at different capacitance values and see if there is an offset that is able to be taken into account. That would be the next step, if the self-made cables with old coaxial cables doesn't work.

My knowledge of the principles behind coaxial cables is limited, but since this is a measurement device and it's being terminated at the clamps or the sample holder in my case, the impedance shouldn't significantly matter? Maybe the calibration feature should be good enough to account for any differences.

I found two identical cables and cut them in half. Now I have the four cables that I needed. I'll try to measure with that soon.

As I'm sure you probably know, the HPot and LPot cables should be connected directly across the test cell. What you're doing is putting a calibrated AC current through the cell (with the HCur and LCur leads) and measuring the voltage across the cell (with the HPot and LPot leads) to determine it's complex impedance (their ratio, in magnitude and phase).

The OP said:

I just have a short length of coaxial cable that is soldered directly to my measurement device.

==

RG-58 has a capacitance of 28.3 pf/foot. That's like 3.5 pf for every 3 inches.

http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=5&cad=rja&ved=0CFoQFjAE&url=http%3A%2F%2Fwww.epanorama.net%2Fdocuments%2Fwiring%2Fcoaxcable.html&ei=ZMTeUs2AJcO-sQTSxoLQDA&usg=AFQjCNFziTHUR22PA8N0aytUUqSRR0HGuQ&sig2=tWc5cehVMtJ1m2EyFsNTCA&bvm=bv.59568121,d.cWc

You really do have to figure out how to compensate the fixture too. Two parallel plates is a capacitor. Therefore the open test, should be with nothing between the plates. The short test should be with a solid block between the plates about the same thickness,

Your fixture is a capacitance too: http://hyperphysics.phy-astr.gsu.edu/hbase/electric/pplate.html even without the material your trying to measure.

==

PS: Where do I come from? The semiconductor industry, I had wired an environmental chamber to to do low current measurements with 4 Triax connections. I did have have trouble doing measurements in the pA range due to the triboelectric effect. Wire movement introduces current. Inside, since I could not find a small amount of high temp coax, I used twisted pair. The other pair was used as a guard and wasn't connected. The devices were placed on a guarded plate.

Yeah, I realize the cable would have some capacitance, but that should be constant and easy to compensate for. What I didn't realize is that it seems to have caused other problems with the meter. Thank you for the links, I will review those and use them to formulate a more scientific approach. Now it's working well enough, that with some basic research I can possibly get a short paper out of this.

I made the new cable as shown above. I could only find these alligator clips with banana plug inputs, so I had to make my do with that. It's not ideal because now there's a short length of unshielded wire (and clamp) but in an evening after everyone was already gone, that's the best I could do :) What you can't see is the test holder that has only about 2 inches of coaxial cable (just so there is something to clamp to). I wanted no coaxial cable on it to reduce cable length, but this is good enough. I twisted the pairs to hopefully reduce induced noise. If that's even a factor, I don't know :)

It reads very stable values, but sometimes the readings to haywire. I can't figure out what is going on, so maybe there's a machine nearby

Winfried, the coaxial cable seems to work rather well, despite the wrong impedance. I honestly don't know if the wrong impedance is causing other errors. The calibration feature works, at least superficially.

Also, now it doesn't read beyond -90 degrees. It reads a phase of about -86 degrees which would make better sense.

Thank you everyone for the replies. I don't think I would have figured out a solution so quickly without your help.