[sci.electronics] measuring a capasitance?

s32852b@puukko.hut.fi (Petri 'Pete' Karha) (03/04/91)

I got a really tricky thing to do and I would appreciate some help. We
must design a device that can measure a varying capasitance the value of
which goes from 0 to 100 pF. Well that's not tricky, just a bridge but
it should have about 0.1% accuracy and it should work in a wide
temperature range (about -20-+70 C). Any good ideas?

--
s32852b@puukko.hut.fi    ! When the mail-daemon calls in the morning,
petek@otax.tky.hut.fi    ! There's a note upon my ~/.forward,
Phone: (Work)   5118 293 ! Just to bring my mail to the /dev/null,
       (Home)   8053 655 ! 'cause I don't live here no more. -J.Burnette&me.

whit@milton.u.washington.edu (John Whitmore) (03/05/91)

In article <1991Mar4.085406.25128@santra.uucp> s32852b@puukko.hut.fi (Petri 'Pete' Karha) writes:
>I got a really tricky thing to do and I would appreciate some help. We
>must design a device that can measure a varying capasitance the value of
>which goes from 0 to 100 pF. Well that's not tricky, just a bridge but
>it should have about 0.1% accuracy and it should work in a wide
>temperature range (about -20-+70 C). Any good ideas?

	I don't know how good an idea it is, but think about this
possibility: look for the phase shift vs. frequency of the capacitor
as driven from an oscillator.  Choose some phase shift (like
45 degrees) that can easily be detected.  You will need, of course, 
good control of the oscillator's output impedance (use thermistors 
if you have to).
	Then sweep the frequency until you find the magic phase 
shift; measure that frequency.  The frequency measurement should
be trivial to the accuracy you want; the only other component that
need be temperature-stable is a single resistance (to provide the
output impedance of the oscillator).
	The advantage over a bridge (three critical components,
at least one adjustable) is in reduction of the number of
gizmos that have to be temperature-stable to 0.1%.  

	John Whitmore

	

joeld@hpnmdla.hp.com (Joel Dunsmore) (03/20/91)

In sci.electronics, whit@milton.u.washington.edu (John Whitmore) writes:

>    	I don't know how good an idea it is, but think about this
>    possibility: look for the phase shift vs. frequency of the capacitor
>    as driven from an oscillator.  Choose some phase shift (like
>    45 degrees) that can easily be detected.  You will need, of course, 
>    good control of the oscillator's output impedance (use thermistors 
>    if you have to).

    Assume Zo=50 ohms, C=50pf, then F(@45 deg)=63.66MHz.

    Now, with a 1 degree error, C=51.78pf to 48.28 pf. So.. How hard
    is it to get much better than 1 degree?  Ans. Hard.

    At .1 % Capacitance error, your phase error would need to be
    0.03 degrees.


Also, assuming no math errors on my part.

Joeld.

whit@milton.u.washington.edu (John Whitmore) (03/23/91)

In article <7380016@hpnmdla.hp.com> joeld@hpnmdla.hp.com (Joel Dunsmore) writes:
>In sci.electronics, whit@milton.u.washington.edu (John Whitmore) writes:
	{responding to question about capacitance meter}
>
>>    	I don't know how good an idea it is, but think about this
>>    possibility: look for the phase shift vs. frequency of the capacitor
>>    as driven from an oscillator.  Choose some phase shift (like
>>    45 degrees) that can easily be detected.  You will need, of course, 
>>    good control of the oscillator's output impedance (use thermistors 
>>    if you have to).
>
>    Assume Zo=50 ohms, C=50pf, then F(@45 deg)=63.66MHz.
>
>    Now, with a 1 degree error, C=51.78pf to 48.28 pf. So.. How hard
>    is it to get much better than 1 degree?  Ans. Hard.
>
>    At .1 % Capacitance error, your phase error would need to be
>    0.03 degrees.

	I was not assuming phase MEASUREMENT, just phase COMPARISON
(i.e. I was assuming one would use a phase-locked loop).  So, the
reasonable phase error for a PLL is the relevant specification.
My source on this topic assures me that with a good loop filter,
phase noise can be held to 1mrad without too much trouble.  That
corresponds to 0.05 degrees, which is nearly what you calculate.
	If one was going to do phase measurement for some
arbitrary phase shift, the difficulties you speak of would,
of course, kill the scheme. 

	I was assuming the device would work at considerably lower
frequencies than 63 MHz; I would have chosen something like 50 kOhms
for the output impedance, and in the few-kiloHertz frequency range
CMOS can probably handle the task.  I think the RLC digi-bridge we use
gets good accuracy from a 1kHz fixed test frequency.

	John Whitmore

tomb@hplsla.HP.COM (Tom Bruhns) (03/26/91)

s32852b@puukko.hut.fi (Petri 'Pete' Karha) writes:
>I got a really tricky thing to do and I would appreciate some help. We
>must design a device that can measure a varying capasitance the value of
>which goes from 0 to 100 pF. Well that's not tricky, just a bridge but
>it should have about 0.1% accuracy and it should work in a wide
>temperature range (about -20-+70 C). Any good ideas?

Well, I'm surprised the only other response you got (that I've seen
here) was a discussion about measurements based on phase shifts.
It seems to me (without many calculations to back it up) that 
measurement of resonant frequency would be a good alternative.  You
could build a very compact oscillator in an oven, one frequency-
determining component of which is the capacitance you wish to measure.
Measuring frequency is usually pretty easy, and the biggest trick 
would be to keep the oscillator in calibration.  "Autocal" features
might be a way to do that.  I assume you mean about .1 pF accuracy,
not .1% of the capacitance; that would be an impossible task.  Indeed,
you will find the capactiance of connecting wires will change easily
by that amount due to humidity changes when you are down in the
single-digit pF range.

Another thought:  quadrupole mass spectrometers typically rely on
extremely stable capacitors to monitor the RF voltage applied to the
quadrupole structure.  If the capacitance you wish to measure can
stand moderate voltages, you might be able to turn that idea around and
compare the current through your capacitance to the current through
a stable reference capacitor; such a ratiometric measurement might
be capable of 0.1pf reliably.  You want to use a moderate voltage so 
accurate rectification is easy.  Actually you are dealing with a
large enough capacitance that you could use a low enough frequency
excitation that rectification might be pretty easy just using an
active (op-amp based) rectifier; mass spec works in the megahertz
region, where it's tough to get the required accuracy.  As a point
of comparison, the mass specs I worked on used 3 pF capacitors to
sample up to a couple kilovolts, and wanted to control that voltage
to a few millivolts stability.  Your problem is at least an order
of magnitude easier; again, temperature control may be required
(an NPO reference cap may have 30 ppm/degree tempco; +/-45 degrees
could put you out of spec...)

tomb@hplsla.HP.COM (Tom Bruhns) (03/27/91)

s32852b@puukko.hut.fi (Petri 'Pete' Karha) writes:

>I got a really tricky thing to do and I would appreciate some help. We
>must design a device that can measure a varying capasitance the value of
>which goes from 0 to 100 pF. Well that's not tricky, just a bridge but
>it should have about 0.1% accuracy and it should work in a wide
>temperature range (about -20-+70 C). Any good ideas?

In my previous posting, I forgot to ask about parasitic elements in this
capacitance.  Are you trying to measure the capacitance in the presence
of a (presumably variable) series or shunt resistance?  If so, that may
temper how you want to go about it.  For example, the shunt resistance
in a tuned circuit would have little effect on the point of resonance, but
in the "compare the currents" method I suggested, could be an insurmountable
difficulty (because the excitation frequency is generally low enough that
the capacitive reactance is itself rather high; that's fairly important to
porper operation of that scheme).  On the other hand, a series resistance 
(so long as it is very much lower in impedance magnitude than the capacitance)
would have very, very little effect on the net impedance (and therefore the 
measurement) because the net impedance is a sqrt of sum of squares thing.  
In response to the ideas, any better description of just what you are doing?  
(Don't want to be nosey, just want to be able to offer more enlightened 
suggestions).