[AMPS] Ceramic transmitting capacitors

[w8nf@mailroom.com]

This was prompted by the dicsussion of silver micas, doorknobs, and
other capacitors in reference to the 1.5kW filters.

The topic of capacitor dielectrics is an extraorinarily complex one. 
Dielectrics can cause grief in more ways than you'd ever realize.  Just
glance at the audio publications and the issues they have with
dielectric absorption, dissipation factor, and all those other non
linear loss mechanisms.  Dielectrics even have hysteresis!  

For RF, the world gets a little bit simpler, but only because we really
can't use any of the low-cost options.  Tantalum, electrolytic,
semiconductor dielectric (those ceramic disk caps you see at 10 for a
penny are actually bipolar transistors with the base unconnected -
they're lousy in any application I can think of, but they're cheap and
fulfil certain regulatory requirements) are all off-limits for us.  We
have to stick with the high-priced spreads!

Simplisticly, in ceramic RF capacitors, you can tell the fundamental
characteristics by the temperature coefficient.  C0G, also known as
NP0, is the so-called holy grail "zero temperature coefficient"
dielectric.  The actual rating is +/-30ppm per degree C.  This is a
relatively low loss dielectric.  It's also low permittivity. 
Therefore, a capacitor using this dieletric must be larger (larger
plate area) than a capacitor using a higher K dielectric.  Higher K
dielectrics result in a negative temperature coefficient.  You will
often see this printed on a capacitor, such as "N4700", meaning
negative 4700 ppm per degree C.  Such a capacitor will have higher
capacitance in the same size package as a C0G, but will also have lower
Q, hence, higher loss.  The famed ATC 100E series of ceramic chip caps
is P90 - positive 90 ppm per degree C.  Even lower loss than C0G.  This
series of dielectrics is all made from various formulations of ceramic,
I believe barium-doped.  I'm no physicist, so actual discussions of
formulation are outside my knowledge area.

Loss is doubly damaging.   First, high Q circuits with high loss
components lose an amazing amount of power.  In 100 - 200 W solid state
PAs, it's common to see P90 or even P150 dielectrics in all tuned
circuits, C0G in bypass components, and sometimes N750 used in coupling
elements.  And second, at the power levels we like to play with, loss
in a small component such as a capacitor leads to heat, which can
rapidly destroy the component.  A small ceramic capacitor really can't
dissipate more than one watt of power without frying itself.

Whether it's a wire-leaded ceramic disk or a doorknob, the same
dielectric characteristics apply.  X7R and Z5U are different families
of ceramic completely and exhibit significantly higher loss.  I do
everything I can to not use them in RF power amps.  X7R, Z5U and many
of the other low-cost dielectrics are chemically quite different from
the "C0G family", have seriously non-linear temperature coefficients,
change capacitance with applied voltage, humidity, pressure, and the
higher-K versions are even microphonic.  It's everything you don't want
in a capacitor, unless you're just interested in bypassing an AC line
with them.

So, what's different between, say, a wire-leaded 1kV 4700pF N750, 1000V
disc ceramic, and a doorknob with the same ratings?  The doorknob,
being larger, is most likely built with larger plates, and greater
spacing.  The mere fact that it is physically larger implies that it
can tolerate more heating, even if the actual dB loss through it is the
same as the leaded ceramic type.

Based on impedance meter measurements I've made, the silver mica
capacitors appear to be very close to the C0G in both temperature
coefficient and loss, but they are slightly larger.  Because of the
larger size, I expect they'll survive higher-current applications
better than leaded ceramic disks.  The unclad metal/mica capacitors by
Unelco and Semco are standard silver micas without encapsulation.

To test capacitors' abilities to handle my application, I abuse-test
them.  I put them in series with a 50 ohm dummy load, and run RF
current through them.  If they're so small in value that they raise the
SWR to the point where my test amp is unhappy, I resonate them with a
series inductor (1/4" refrigerator tubing).  This raises the voltage
across the cap, and the RF current through it is equal to the current
into the 50 ohm load.  I never find a need to voltage test: a 500V cap
seems to work fine to that voltage.  

If I have to test at higher current than my amp will deliver, I'll
create a parallel resonant circuit to purposefully cause circulating
current.  For example, let's say I have a 2200pF cap, and in my circuit
it has to handle 40 amps RF.  If I send 4 amps of RF to my dummy load
and create a parallel resonant circuit with that cap and an inductor of
Q=10, then I'm set.  The capacitor has an Xc of -5 ohms at 14.45 MHz. 
So I'll wind a "square" inductor out of the thickest copper tubing I
can comfortably handle to resonate that cap at 14.45 MHz.  Put them in
parallel across my 50 ohm test fixture and drive 4amps of RF into the
dummy load at 14.45MHz.  That's 800 watts.  Notice that I could have
done it at 28.9 MHz with 200 watts, but I'd have had to create a
parallel network with a Q of 20.  At that level, I'd worry about being
able to make a "good enough" inductor.  After an hour of operation, if
I can barely feel the heat in the cap, I figure it's OK as long as it's
going to get some cooling when operating.  If I think the cap is risky
to begin with, I don't start the test at 800 watts.

Most of us who've done commercial amps have had to make capacitors.  RG
8 works, but is huge, and you have to carefully treat the open end to
avoid corona.  I've made sandwiches out of thin aluminum and 1/32"
teflon PC board material, but they end up temperature sensitive and
large, but they'll handle lots of current.  For HV screen bypass, 3M
Pyralux (polyimide-based PC material) is great.  It etches like PC
board material, but is thin enough to create decent capacitance, and is
available with pretty high voltage ratings.

Paralleling works as expected for increasing current capacity.  Series
connections are a bit more dodgy; the RF voltages will distribute
according to the capacitor reactances, but since we've intentionally
chosen low-loss capacitors, static build-up is a genuine possibility.  

A former colleague, who designed 1 MW and up transmitters for
Continental in Dallas, told me that they pretty much design and build
all of their own matching components above the 20kW level.

Regards,

Dave W8NF

---------------------------------------------------------------
This Message was Powered by Xcel Communications
Sign up for your FREE EMAIL account today at http://www.mailroom.com
Give your FAX machine an email address http://www.faxroom.com


--
FAQ on WWW:               http://www.contesting.com/FAQ/amps
Submissions:              amps@contesting.com
Administrative requests:  amps-REQUEST@contesting.com
Problems:                 owner-amps@contesting.com