Capacitor DA test, ala Walt Jung

[jneutron]

Posted these questions on another forum..in essence, the responses were not nice....Steve Eddy recommended that I try posting here as both Scott Wurcur and Walt Jung post here....this is the post in it's entirety.

Start:

Thanks again, John Curl, for the e-file..
Some questions..

In all the scope photo's but one, the Cref is listed as polypropylene, but I note that the values change based on the size of Ctest.

You show:

8 uF/200 V
8 uF/800 V
1 uF/200 V
.1 uF/200 V
and a 1uF/50 volt Teflon.

First Q...Did you do a test comparison between same capacitance polypro's with only the Ctest voltage rating changed? That would almost show the variation in errors caused by the electric gradients in the dielectric. I say almost, because to get the higher voltage, they have to change the dielectric thickness, then increase the effective plate area to compensate..How they get that additional plate area will affect both esl and esr, as well as possibly making the unit more prone to skin effect based capacitance drop, depending on construction technique used and lead exit strategy.(hee hee...had to get that phrase in there somehow..).

Q2: Did you compare save voltage polypro's with different capacitances? That would look specifically for the geometry based skinning of the unit..As part of that, possibly just paralleling two polypro's of the same value as Cref, to double the test leg capacitance and compare to a larger Ctest.

Q3: did you try baselining the rig with a pair of identical polypro's for Cref and Ctest?

Q4: How did you test the summing bridge to make sure that the error presented to it did not cause any input saturation based effects that generated a recovery period..From my earlier days of test using this type of equipment, I always had to watch out for the amp capability to recover from input "overload", usually a speed/amplitude phenomena. Sometimes, I had to put a front end resistor fed diode bridge, to very quickly limit the output drive being sent to the input stage..into a 50 ohm coax, that style of bridge gave me clean sub nanosecond truncation and recovery, alowing the settling time to be eliminated.

Q5: Has anyone updated this test setup using this generation of IA's, to make the circuit useable for capacitance values in the range of wires?? I can see the Cref needed already..a pair of copper pipes, ten feet long, arranged coaxially, and spaced every 2 feet so often with a teflon ring.


I realize that publishing an article like this limits the authors, sometimes these types of issues get dropped to the cutting room floor. So I would guess that some of my questions have been addressed, some perhaps not...

But certainly worthy of an update, don't ya think?

END


I am interested in understanding how capacitor geometry will affect it's response in this test..

Also, I am interested in souping it up to be able to check for possible DA issues on wires..understanding that the circuit is probably not capable of reading low nanofarad high pico capacitance, and will probably require a higher speed IA.

Cheers, John

[SY]

Quick Qs:

I haven't seen the HFNRR version of this paper, just the Audio Amateur one. Did the HFNRR version have the same 1000x typo error on the Teflon vs polypropylene photo?

Are there more suitable instrumentation amps available these days for the sorts of wire tests you're interested in doing?

[SY]

One more quick Q: do you think that this test could be done better these days by taking the data "single-ended" (as it were), then doing a digital subtraction, rather than using an instrumentation amp?

[jneutron]

Originally posted by SY
One more quick Q: do you think that this test could be done better these days by taking the data "single-ended" (as it were), then doing a digital subtraction, rather than using an instrumentation amp?

I guess it would be very good for all but very deep nulls..I'd worry aout resolution issues.

The origional app with the IA chosen points out some very good differences between cap types..

I do have a problem, however, with the procedure...they use a reference capacitor, say a teflon or polypropylene, and adjust some resistors to null the bridge. But, what exactly was nulled?

For identical capacitors, yes, the null is reducing the residue...but for different types, say an electrolytic...what is being nulled?

A high value electrolytic has a capacitance dropoff with frequency. Some can be about half the effective value at 20Khz.. Since the input of the test system is being fed via a 8 us RC filter, what is the actual capacitance value during the leading edge of the pulse? If the capacitance is half, is the bridge nulled for that value? Or, does one wait for the amount of time necessary for the rest of the capacitor to be attached to the circuit?

In essence, how does the skin effect inherent in larger electrolytics get compensated for in this rig? Can it be?

What model would be required to include skin effect inductance change and distributed capacitance? sounds rather complex..

What is the skinning vs value tradeoff.. How does the pickoff points in the foil construction affect the skinning?

The confounding issue with skinning within a capacitor is the fact that as frequency rises, the self inductance of the foils reduces, and so does the total area of the capacitor that is actually passing current..Normally, ESL is described as a series component that remains constant..

As for using the rig to test wire DA....It's too slow for the cap values we'd look at..but I really do like your digital capture possibility...For Curl and Jung, I don't think it was an option for them..what they did was far and above what was known at the time, using a nicely creative circuit idea.

As for the electrolytics...it is certainly not clear that DA is in fact the entity that was being seen, as skinning would confound the results..

Unfortunately, I don't think that at that time ('86), the standard model of a capacitor included skin effect as an element, so standard ESL numbers would have been considered.

Perhaps revisiting the test is in order, armed with new knowledge..

Cheers, John

[SY]

Well, I would think that resolution would not be a huge issue, given the rather sophisticated DAQs available these days at surprisingly low cost.

RE: what value one should choose to null for electrolytics, my feeling is that the funny effects of frequency ought NOT to be nulled out. Let's look at all errors, then break down the error signal into different components, i.e., error from dC/df, loss tangent, DA, whatever.

What I would find most interesting is to actually put a coupling cap into an appropriate load, apply music or test signals of appropriate bandwidth and magnitude to the circuit, then differentially measure the residual signal across the cap. You then have the option of ratioing the error to the theoretical error due to the ideal network response and seeing what the cap is doing wrong. Or you could compare two caps by ratioing their error signals.

I know, I know, you want to make this work for wire!

[jneutron]

Originally posted by SY
I know, I know, you want to make this work for wire!

Well, yah...but first, lets figure out what we are actually doing with caps..trying to do this with 100 pf of wire without understanding what's happening at the 1uf level would be rather absurd.

As it doesn't appear that all the residue of an electrolytic is a direct result of the dielectric, but rather, some combination of geometry and dielectric. And, even the tef's and polys may have some confounding effects, based on their construction. It would be nice to confirm or refute the possibility that what is seen is strictly dielectric related..

Unfortunately, a two page article doesn't provide enough room to make a really long presentation of all the tests, but only a good summary..

Cheers, John

PS..I was hoping that Jung and Wurcer could drop in..

[SY]

Scott probably will. I've got some interesting cap simulations that he's done with an interesting conclusion- but I'd rather get it from the horse's mouth (so to speak).

[scott wurcer]

I was just a general consultant on these tests. It's been a long time, but I think that if you take a reference capacitor and resistor and assume that they are essentially 'perfect' over a frequency range of interest then the bridge null technique can be used as a very high resolution way to extract the real and imaginary parts of the impedance of the capacitor under test. You could then build a model of the capacitor ala Bob Pease's article on fixing ultra-high resolution integrators. You could then drop this model into your circuit. I'll pass on making any judgements on telling anything about how this relates to the 'sound ' of any particular capacitor.

I have tried to measure the 'Q' of a 10pF electret microphone using this technique and was unable to get anywhere near enough resolution. The mecanical setup was impossible. Since the impedances at audio frequencies of 10pF are so huge I couldn't shield or bootstrap all the strays to the femto-Farad level. Measuring 100pF of wire capacitance would have the same mechanical problems i.e. the bundle of wire has strays all over the place and you can't bootstrap them all out so you could trust the null.

[peranders]

Have you seen this page about the subject?
Bob Pease wrote some....
http://www.national.com/rap/Applicat...570,28,00.html

[scott wurcer]

Originally posted by peranders
Have you seen this page about the subject?
Bob Pease wrote some....
http://www.national.com/rap/Applicat...570,28,00.html

Yes, that's the article that I was talking about. I didn't know it was on line. The distortion of capacitors is another issue, but a real one. Every ADSL modem uses 2-4 expensive film capacitors because the system would not work with cheap ceramic ones.