capacity measurement

[Jeroen Vriesman]

Hi,

I need to measure a capacity of 5..6 pF within 30 uS, at 0.01% accuracy.

The accuracy does not have to be absolute, doesn't have to be completely
lineair also, as long os noise and tempco don't affect the measurement more
than 0.01% it's ok. (within about 15 deg C temperatur range).

I've got some ideas, but does aynone have some realy good suggestions or
links?

Cheers,
Jeroen.

 

[Rene Tschaggelar]

Hi,

I need to measure a capacity of 5..6 pF within 30 uS, at 0.01% accuracy.

The accuracy does not have to be absolute, doesn't have to be completely
lineair also, as long os noise and tempco don't affect the measurement
more than 0.01% it's ok. (within about 15 deg C temperatur range).

I've got some ideas, but does aynone have some realy good suggestions or
links?

Yes, take a sweeper and measure the impedance of the series LC
being built from your unknown C and a known L. The 100ppm accuracy
may not be achievable.
Well you'll have to calibrate your LC with a known C once.

Rene

 

[Phil Hobbs]

Yes, take a sweeper and measure the impedance of the series LC
being built from your unknown C and a known L. The 100ppm accuracy
may not be achievable.
Well you'll have to calibrate your LC with a known C once.

Capacitive gauges do at least this well. You do it by putting your
capacitor in a Wheatstone bridge with another cap, and driving it with a
transformer at around 100 kHz. This is much better than you can do with
resonances. (Capacitive gauges haven't been made using tank circuits
since the 1950s.)

Cheers,

Phil Hobbs

 

[John Larkin]

Capacitive gauges do at least this well. You do it by putting your
capacitor in a Wheatstone bridge with another cap, and driving it with a
transformer at around 100 kHz. This is much better than you can do with
resonances. (Capacitive gauges haven't been made using tank circuits
since the 1950s.)

Except, of course, this one...

http://www.aade.com/lcmeter.htm

which is an extremely cool gadget.

John

 

[Ben Bradley]

Hi,

I need to measure a capacity of 5..6 pF within 30 uS, at 0.01% accuracy.

It's a two-step process:

1. Use the cap as part of the frequency-determining circuit of a
high-frequency oscillator (let's say 300 MHz). Count cycles of the
oscillator output for 30 uS, and do a table lookup of the count vs.
(pre-tested, calibrated) capacitance. For 0.01 percent accuracy, you
need a table with at least 10,000 entries, but current PC's have
plenty of RAM.

2. Pay me a consultation fee.

 

[Bill Sloman]

Hi,

I need to measure a capacity of 5..6 pF within 30 uS, at 0.01% accuracy.

The accuracy does not have to be absolute, doesn't have to be completely
lineair also, as long os noise and tempco don't affect the measurement more
than 0.01% it's ok. (within about 15 deg C temperatur range).

I've got some ideas, but does aynone have some realy good suggestions or
links?

Hmm. 0.01% is a bit less than 14-bit accuracy. Analog Devices and
Texas Instrument have both got 16-bit A/D converters with conversion
times around one to two microseconds, and apperture times which are a
whole lot shorter

http://www.analog.com/UploadedFiles/Data_Sheets/31038630530425AD7653_a.pdf

for the Analog Devices AD7653 gives an acquisition time of 0.25usec.

I'd make myself a a very stable 5uA constant current source, and
direct it into the commoned sources of a pair of SD214 lateral
MOSFETs. Switch this into your capacitance for 1usec to produce a
roughly 1V step. Buffer the voltage across the capacitor with a fast
FET-input op amp - the OPA655 comes to mind - and digitise the op amp
output. Use a third SD214 to discharge the capacitor to ground (or
some other stable voltage) for, say, 1usec. Turn off SD214, and wait
for the capacitor voltage to stabilise again, then digitise the
starting voltage.

This connects two SD214 aysmmetrical lateral MOSFET transistors to the
capacitor, adding about 1.5pF per transistor to the capacitance being
measured,

http://www.linearsystems.com/datasheets/SD214.pdf

while the OPA655 adds another 1pF

http://focus.ti.com/lit/ds/symlink/opa655.pdf

You'd probably need to use the circuit to measure just the capacitance
of these parts as a function of temperature (at half the current) to
have any hope of maintaining 0.01% accuracy on the capacitance you
want to measure.

The job seems to be do-able, but the 0.01% might be difficult to
achieve.

It would be easier if you could find a 5.6pF capacitor that was stable
to better than 0.01% over the temperature range that you want to
cover, but
+/-30ppm/C ceramic disks would seem to be the best you could get and
that is +/-0.01% in +/-3C.

 

[Winfield Hill]

Jeroen Vriesman asked...

I need to measure a capacity of 5..6 pF within 30 uS, at 0.01% accuracy.

The accuracy does not have to be absolute, doesn't have to be completely
linear, as long as noise and tempco (over 15 deg C range) don't affect
the measurement more than 0.01%

Phil Hobbs answered:

Capacitive gauges do at least this well. You do it by putting your
capacitor in a Wheatstone bridge with another cap, and driving it
with a transformer at around 100 kHz. This is much better than you
can do with resonances. (Capacitive gauges haven't been made using
tank circuits since the 1950s.)

John Larkin noted...

Except, of course, this one... http://www.aade.com/lcmeter.htm
which is an extremely cool gadget.

0.01% of 5pF is 0.5fF, and this is 200x better than the cool AADE
meter can achieve. But happily it's a 10x easier spec than the
performance of my servo-balanced capacitance bridge design, which
circuit I've posted here several times. However my circuit runs
at 10kHz and probably takes over 5ms to reach 0.01% settling after
a step capacitance change. Perhaps if my circuit were modified to
operate at say 500kHz it could achieve a 30us 100ppm step response,
while not suffering a performance degradation beyond 100ppm.

Thanks,
- Win

whill_at_picovolt-dot-com

 

[Ian Buckner]

accuracy.

It's a two-step process:

1. Use the cap as part of the frequency-determining circuit of a
high-frequency oscillator (let's say 300 MHz). Count cycles of the
oscillator output for 30 uS, and do a table lookup of the count vs.
(pre-tested, calibrated) capacitance. For 0.01 percent accuracy, you
need a table with at least 10,000 entries, but current PC's have
plenty of RAM.

2. Pay me a consultation fee.

You can use a lower frequency, gate an integer number of cycles into
a computing counter and measure the period. You can get 12 digits per
second of observation time, so 30usec gets a resolution of 0.3ppm.

Regards
Ian

 

[Tim Shoppa]

Hi,

I need to measure a capacity of 5..6 pF within 30 uS, at 0.01% accuracy.

The accuracy does not have to be absolute, doesn't have to be completely
lineair also, as long os noise and tempco don't affect the measurement more
than 0.01% it's ok. (within about 15 deg C temperatur range).

Accuracy is different than precision. Most of the other solutions so
far suggested might deliver that much precision, but as
a practical matter getting even 2% accuracy will require a bridge
configuration.

20% accuracy will require careful elimination of strays.

What sort of NBS traceability do you need?

Tim.

 

[Jeroen Vriesman]

Nonlinearity is not a problem, neither is absolute precision, it's more
about stability.

I'm thinking about how to design the input stage, just using a current
source and measure the time it takes to reach a certain voltage would be
ok, but the input of a high speed opamp is at least 1pf, and varies with
temperature.

If I first divide the signal with two resistors, then a high smd resistor
has about 0.2 pF

Does anyone know the tempco of opamp input capacity? Datasheets don't
specify it.

Cheers,
Jeroen.

 

[Frank Miles]

Nonlinearity is not a problem, neither is absolute precision, it's more
about stability.

I'm thinking about how to design the input stage, just using a current
source and measure the time it takes to reach a certain voltage would be
ok, but the input of a high speed opamp is at least 1pf, and varies with
temperature.

If I first divide the signal with two resistors, then a high smd resistor
has about 0.2 pF

Does anyone know the tempco of opamp input capacity? Datasheets don't
specify it.

You should probably bootstrap the power supplies for the opamp. With
proper gain the opamp C variation should be small enough to ignore.
And of course avoid CMOS op amps, which generally have high Cin.

-frank

[snip]
--

 

[Jeroen Vriesman]

Hi

Schematic attached,

if R1 has 0.2 pF, and the input of the opamp is 1pF, then a 10% variation
is opamp input capacity is extra/less 0.05%, which is too much.

If I replace R1 by 5 resistors is series, it would be ok for a 10% opamp
input capacity variation, but any idea how much it actually is? 5%, 10 50%?
I've got no idea.

Cheers,
Jeroen.

Nonlinearity is not a problem, neither is absolute precision, it's more
about stability.

I'm thinking about how to design the input stage, just using a current
source and measure the time it takes to reach a certain voltage would be
ok, but the input of a high speed opamp is at least 1pf, and varies with
temperature.

If I first divide the signal with two resistors, then a high smd
resistor has about 0.2 pF

Does anyone know the tempco of opamp input capacity? Datasheets don't
specify it.

You should probably bootstrap the power supplies for the opamp. With
proper gain the opamp C variation should be small enough to ignore.
And of course avoid CMOS op amps, which generally have high Cin.

-frank

[snip]

 

[Phil Hobbs]

Hi

Schematic attached,

if R1 has 0.2 pF, and the input of the opamp is 1pF, then a 10% variation
is opamp input capacity is extra/less 0.05%, which is too much.

If I replace R1 by 5 resistors is series, it would be ok for a 10% opamp
input capacity variation, but any idea how much it actually is? 5%, 10 50%?
I've got no idea.

Cheers,
Jeroen.

Transformers, especially transmission-line transformers, are extremely
stable with time and temperature. Why not profit by the painful
experience of people who do this for a living? (I'm not one of them,
but I go to their parties.)

Cheers,

Phil Hobbs

 

[Frank Miles]

-=-=-=-=-=-
[snip]

I'm thinking about how to design the input stage, just using a current
source and measure the time it takes to reach a certain voltage would be
ok, but the input of a high speed opamp is at least 1pf, and varies with
temperature.

If I first divide the signal with two resistors, then a high smd
resistor has about 0.2 pF

Does anyone know the tempco of opamp input capacity? Datasheets don't
specify it.

You should probably bootstrap the power supplies for the opamp. With
proper gain the opamp C variation should be small enough to ignore.
And of course avoid CMOS op amps, which generally have high Cin.

-frank

[snip]

Sorry I don't recall why you need the input resistors. I'll try to
modify your drawing to show what I mean by bootstrapping. Sorry
the drawing is not up to Win's high standards. The 'n' transistor
is NPN; 'p; is PNP. Of course, if you absolutely have to have more
voltage gain in this stage you could resistively divide the output
to drive each supply pin -- keeping the silicon moving along with the
input signal. If the signal is small, driving the negative (substrate)
might be sufficient.

-frank

+|
\ | resistor to +
-=-=-=-=-=- n \|___|
/| |
/ | ---
|/ / \ zener
current source | ---
R1 |\| |
----|------/\/\---|---------| \ |
| | | >---**--
_|_^ | ----| / | |
_/_ | | |/ --- |
/| | | | / \ zener
| | | \| --- |
| | | p |---| |
| | | /| | |
| | | -V | |
| | ---------------|
| | |
| \ resistor to -
| /
| \
| |
| |
---|-------------|-----|-------------------

-=-=-=-=-=-

--

 

[Jeroen Vriesman]

Hi,

couldm't really read your schematic, but I know what you mean by
bootstrapping, I just don't see how that would do anything about the opamp
input capacity. I need the input resistors to minimize the influence of the
input capacity.

Today I tried something with transmission line transformers, looks
promising, was a suggestion of Phil Hobbs.

Cheers,
Jeroen.

-=-=-=-=-=-
[snip]

I'm thinking about how to design the input stage, just using a current
source and measure the time it takes to reach a certain voltage would
be ok, but the input of a high speed opamp is at least 1pf, and varies
with temperature.

If I first divide the signal with two resistors, then a high smd
resistor has about 0.2 pF

Does anyone know the tempco of opamp input capacity? Datasheets don't
specify it.

You should probably bootstrap the power supplies for the opamp. With
proper gain the opamp C variation should be small enough to ignore.
And of course avoid CMOS op amps, which generally have high Cin.

-frank

[snip]

Sorry I don't recall why you need the input resistors. I'll try to
modify your drawing to show what I mean by bootstrapping. Sorry
the drawing is not up to Win's high standards. The 'n' transistor
is NPN; 'p; is PNP. Of course, if you absolutely have to have more
voltage gain in this stage you could resistively divide the output
to drive each supply pin -- keeping the silicon moving along with the
input signal. If the signal is small, driving the negative (substrate)
might be sufficient.

-frank

+|
\ | resistor to +
-=-=-=-=-=- n \|___|
/| |
/ | ---
|/ / \ zener
current source | --- R1 |\| |
----|------/\/\---|---------| \ |
| | | >---**--
_|_^ | ----| / | |
_/_ | | |/ --- |
/| | | | / \ zener
| | | \| --- |
| | | p |---| |
| | | /| | |
| | | -V | |
| | ---------------|
| | |
| \ resistor to -
| / | \ | | |
| ---|-------------|-----|-------------------

-=-=-=-=-=-

 

[Tim Shoppa]

Nonlinearity is not a problem, neither is absolute precision, it's more
about stability.

You start out asking for a 0.01% capacitance measurement, and now you
decide that nonlinearity is fine, and that you don't really need all
that much precision

I think (correct me if I'm wrong) what you're really asking for is
monotonicity. Am I getting close?

A capacitive voltage divider fed by a fixed frequency, measuring the
AC voltage across eithor or both the "known" and "unknown" unit,
might get you started. Having the unknown hooked to ground is perfectly
compatible with this approach. The gross
non-linearities can be corrected after measurement of the voltage,
if that's of any relevance to you.

Tim.

 

[Jeroen Vriesman]

Yes, you are right I didn't state my question properly, I should have
asked for monotonicity and stability to begin with.

But anyway, I got it working to about 300ppm already, your aproach is not
the one that would work I think, whatever I would measure the voltage with
will have an (instable) input capacity, which is the main problem, then
there is the tempco of the other dividing capacitor, and I would have
problems with the high frequencies I need to measure it within 30 us. It's
only about 5.5 +/- 0.5 pF I'm trying to measure.

But anyway, I'm on the way to 100ppm so you can consider this thread
closed, thanks anyway.

cheers,
Jeroen

joenix AT gmx DOT net