C0G ceramics vs humidity

To solve the problem on the right side of the vehicle, you just need some Furrier poles.

Cheers

Phil "unstable anyway" Hobbs

Are oscillations pole dancing?

Only when using gyrators.

Cheers

Phil Hobbs

No, it works fine.

The switch resistance does not matter, it still dumps all the charge within a small fraction of a cycle.

S1/S2 does not really need to be a switch, it can be a driver e.g. FSA4157 as labelled. There is no charge injection effect. You have to be careful with synchronous supply ripple from it but as I said everything can be made ratiometric to the supply anyway so it does not matter.

Charge injection from S3/4 is low and constant because all terminals work at GND. This also eliminates any effect of C3 DA as far as I can tell, it lives at 0V too. C2 does have DA but contributes no long-term error. It should be NPO to get fastest result.

Bottom line is that there is nowhere for the charge to go except via R3. The duty averages to the correct value. A couple of seconds of numerical averaging gives you a result with sub-ppm stability and repeatability linear to around 2ppm, in my tests.

The circuit is insensitive to shunt capacitance on the terminals of Cx. This allows you to use coax cables to connect to a shielded box without the cable capacitance causing an error. The shielded box can have a few capacitors and switches, this is what allows the linearity to be demonstrated to ppm accuracy.

As for "TCs everywhere" the only ones that matter are those of the reference resistor and the reference crystal. Crystals can be very low TC so it is the reference resistor that dominates. A 0.001% resistor is off the shelf. There are no 0.001% capacitors and if there were you could not measure it.

Doesn't that depend on having a reference capacitor with unknown characteristics? Even the high-end commercial bridges seem to struggle to get much better than 0.1% accuracy. But I guess you can have identical capacitors, one in low-RH and one in high-RH.

Oh I forgot the opamp TC. You can use a low-TC opamp, but the real circuit has a micro operating the switches and doing the duty calculation. This lets one measure C2 and then the offset by turning switches off and looking at the residual ramps. You can also measure the charge injection by disabling the driver and then correct for this. E.g. a version with 1000pF range has an effective "offset capacitance" of ~0.1pF which can be measured and subtracted out of the result.

I proposed making an AC Wheatstone bridge using four of the NP0 caps under test, off the same reel. Null the bridge, expose two caps to humidity, and expose two caps to a dessicant. All you measure is the delta-c.

Sure, that works. I was really responding more generally to your comment about a ppm c-meter. I thought you might like the circuit.

I do like it, but I don't think it would work very well to measure the effect of humidity on an NP0 cap. The TC of A1 and R3 will be problems when trying to resolve capacitance to PPMs. A humidity sensitivity might take days to resolve.

I still think that doing a before-and after-boiling measurement of a few caps with a Boonton or the equivalent would put the issue to rest right away, by showing no measurable change even under extreme conditions.

Cheers

Phil Hobbs

A comfortable lab environment should be well within 5 degrees C. Vishay Z201 has a claimed TC of 0.05ppm/K.

A1 will contribute too but it can be a sub-ohm switch/driver so the effect of its drift can be made negligible.

But I guess the circuit is more valuable for absolute measurements.

I like it John D. Say, a silly question about the loop and your comparator A1. Do you set a rather wide hysteresis so that you capture several cycles of the capacitor charging?

It might be nice to try it at a few different frequencies.. Sure it's not a sine wave drive, but still the fundamental is given by your pulse drive.

George H.

(I lost a partial reply somehow so that might end up posted too. Trying Again).

Hi George,

Actually you can ditch the comparator and just wrap the 20k around U1 then you end up with a simple DC voltage output. Low pass filter again to get rid of ripple from the excitation.

The comparator is just a dirt-cheap way to digitise the signal accurately for a microcontroller (or a frequency meter that can calculate duty cycle). It saves a delta-sigma converter (or it *is* a poor mans d-s I suppose). But you can just use a high-end DVM or something. Then again you really want it ratiometric if possible so you would need a good stable reference voltage too.

My final circuit had a microcontroller ADC input as the "comparator" so the limits were set in software, yes wider limits are better I think so as to reduce the number of switching edges. Since the feedback current is sort of "undefined" during these. Although it worked fine regardless really. You program the micro to timestamp the generated edges, then accumulate totals for "time on" and "total time". These are then divided every few seconds to get the result. You can measure an (average) duty cycle *really* accurately like that, ppb. While testing I could dial in a duty of 0.123 456 789 on my arb generator and get that same number out of the microcontroller measurement!

It's all fun stuff, I spent far too long playing with it as you may gather.

I think we just need to change the definition of the "farad" to use my circuit. They did it for the Volt when Josephson Junction references came along. And what's so great about sine waves anyway?

(humor)

Thanks, Too short of an averaging time might lead to problems.

Hey, every once in a while you get to play with your circuits for longer than you should... that's part of the fun.

George H.