Connect with us

RF coupling with multiple capacitors

Discussion in 'Electronic Design' started by Joel Kolstad, Nov 15, 2005.

Scroll to continue with content
  1. Joel Kolstad

    Joel Kolstad Guest

    I've been playing around with trying to find optimal capacitor values to
    parallel for wideband RF coupling, and have found the following:

    -- One capacitor really seems to do pretty well, despite the common practice
    of paralleling a big cap with a little cap. Something like a 1uF 0805
    capacitor, even though it has a self-resonant frequency of ~6.5MHz, appears to
    work quite well up to at least 3GHz. AVX's SpiCap program tells me this
    capacitor has an impedance of about 11.31 ohms at 3GHz, so the magnitude of
    the -j11.31 in series with 50 ohms is only 51.3 ohms, so I suppose this makes
    sense.

    -- Two capacitors in parallel don't seem to interact as you'd expect. That
    is, although simulations (using an ESR-Cap-ESL model of the capacitors) would
    indicate that you get an anti-resonance at frequencies beyond the larger cap's
    SRF (i.e., it appears inductive and, being in series with a smaller cap,
    appears as an open circuit). Cypress has an application note describing this
    'problem:' http://hsi.web.cern.ch/HSI/s-link/devices/g-ldc/decouple.pdf --
    their recommendation being, that if you must parallel caps, use all the same
    values or keep the values as close together as possible). On an actual PCB,
    however, I'm not seeing this behavior... I can place a 1uF cap in parallel
    with a 100pF cap (both bridging two microstrip lines), and I don't observe an
    anti-resonance.

    -- One the other hand... for DC biasing/RF blocking... I've found that placing
    a 220nF inductor in series with a 22nF inductor makes a very nice resonant
    circuit (dropping the S_{21} response of the biased device -- a MMIC
    amplifier -- about 15dB at the resonant frequency!). Interesting...

    In summary, then... it seems that (1) multiple paralleled capacitors for RF
    coupling aren't as necessary as you might initially think, (2) one doesn't
    have to worry too much about anti-resonances if you _do_ end up with multiple
    paralleled caps, and (3) one DOES have to worry about resonances with multiple
    series inductors.

    Comments?

    ---Joel Kolstad
     
  2. xray

    xray Guest

    The book "Experimental Methods in RF Design" Hayward, Campbell, Larkin,
    published by the ARRL has a section called "Bypassing and Decoupling".
    Their conclusions about putting two different values in parallel are not
    good. "The results are terrible!" ... "This behavior is a dramatic
    example of lore that is generally wrong."

    They do recommend using a couple caps of the same value and that a
    single larger cap isn't bad.

    That book changed my view on the subject. Good to see other measurements
    that reach the same conclusions.
     
  3. Joerg

    Joerg Guest

    Hello Joel,

    Best is to measure the caps on an impedance analyzer and study their
    data sheet. The RF behavior is often in some kind of 'family spec'.
    However, that still leaves the chance that someone changes suppliers later.

    I found cap resonances to be less of a problem than I initially thought.
    But at the same time I usually do not specify a cap that is much higher
    in capacitance than is needed for the job.

    Regards, Joerg
     
  4. Joerg

    Joerg Guest

    Hello Joel,

    Best is to measure the caps on an impedance analyzer and study their
    data sheet. The RF behavior is often in some kind of 'family spec'.
    However, that still leaves the chance that someone changes suppliers later.

    I found cap resonances to be less of a problem than I initially thought.
    But at the same time I usually do not specify a cap that is much higher
    in capacitance than is needed for the job.

    Regards, Joerg
     
  5. Joel Kolstad

    Joel Kolstad Guest

    Hi Xray,

    (Looking at the book...) Hmm... I believe they calculated (simulated) figure
    2.90 and what I'm finding is that, in actuality, the anti-resonance isn't as
    bad as a simulation would suggest. (Although I'm looking at caps as coupling
    elements, whereas they're --iniitally -- looking at bypassing, and they do
    say, "A blocking capacitor is not as critical as a bypass, for the impedance
    on either side will usually be higher than that of the block.") I'm going to
    try out some different values, though -- hopefully I can get results as 'bad'
    as simulation would suggest. :) (Maybe I got lucky in initially finding two
    series inductors that very clearly didn't work well together?)

    ---Joel
     
  6. Joel Kolstad

    Joel Kolstad Guest

    Hi Joerg,

    AVX gives you their 'SPIcap' programs that spits our ESR, ESL, after you
    select package size, dielectric, voltage, and capacitance,. Although their
    program varies ESR as a function of frequency, they fix the ESL based on the
    package size... I'd guess they took a bunch of data, computed a mean value of
    ESL to get the SRF correct, and then varied ESR as a function of frequency to
    fit the data as best as they could (the fact that they use the same ESL for
    0402 and 0603 packages concerns me a little, however!)
    Good to hear you had similar results!
    For inter-stage coupling, I initially thought one ought to aim for something
    like 1/10th or 1/20th the design impedance (e.g., for a 50 ohm RF design, aim
    for 2.5-5 ohms), but having now played with it some, shooting for more like
    1/2 or 1/4 seems more realistic, and will allow for smaller caps and hence
    better high frequency performance.

    ---Joel
     
  7. John  Larkin

    John Larkin Guest

    The important thing is to have the cap width match the trace width,
    and to have a smallish gap in the trace, maybe not the ideal smt
    footprint. For a wider trace, like 110 mills (50 ohms on thick FR-4)
    you can bridge the gap with several identical 0603 or 0402 caps, which
    gives a very nice broadband DC block.

    I tried the DiLabs super-wideband DC block caps, and they didn't seem
    any better than a couple of cheap ceramic caps.

    For really wideband, make a z-cut in the trace and bridge it with a
    mess of ceramic and tantalum caps!

    Most coupling and bypassing lore is plain wrong, especially if it's
    from a capacitor company. People insist that caps shouldn't be used
    above their SRF, which is nonsense; all that matters is their actual
    impedance.

    The "opposite" problem, how to use multiple inductors to make a clean
    broadband DC block, is more difficult. We did disassemble a PSPL
    wideband bias tee, which turned out to be fairly weird.


    John
     
  8. Phil Hobbs

    Phil Hobbs Guest

    You said it. In my first engineering job, many many moons ago, we used
    a lot of 0.1 uF stacked film caps for bypasses. There was one school of
    thought that said that film caps were bad because they became inductive
    at high frequencies and led to lots of ringing. Being naive, I thought
    that I could settle a religious war using data--silly me. So I got a
    few of them, soldered them across the ends of short pieces of hardline,
    and put them on a nice HP vector network analyzer. Result? no
    noticeable self-resonance in any of them--Qs of maybe 0.3 or less--and
    around 100 MHz they went very slightly inductive, about j50 milliohms
    worth--probably indistinguishable from zero ohms on that analyzer. You
    couldn't ask for a nicer bypass cap.

    I sent a memo round to this effect, with the analyzer trace, and was
    told by a senior engineer, "You see? It went inductive. That means
    it'll ring."

    I went back to my cube and forgot about it. About half the boards in
    that system went out with useless 1000 pF ceramics in parallel with
    every one of those nice film caps.

    Cheers,

    Phil Hobbs
     
  9. Joel Kolstad

    Joel Kolstad Guest

    Thanks for the comments, John -- I've been impressed by the various pictures
    of your equipment you occasionally post to ABSE, and I expect the designs
    perform as well as they look.
    Sounds nice. I finally managed to see a couple dB dip in S21 by paralleling a
    2.2uF 0805 cap with a 22pF 0603 cap... although its effect was actually not
    that much larger than a 'divet' I get due to the trace widening from ~55mils
    (this is a .031" board) to nearly triple that where the caps are mounting (the
    original idea was to try mounting up to three caps side by side).
    I'm not quite sure I follow what shape a 'z-cut' looks like?
    Yeah, so I've been 'discovering!'
    Was it something like http://www.mini-circuits.com/dg03-222.pdf ?
    100kHz-4.2GHz for the JEBT-4R2GW-- pretty nice!

    ---Joel
     
  10. Joel Kolstad

    Joel Kolstad Guest

    Phil,

    I'm not quite THAT young, but almost. :)

    I'm liking the idea of a 100nF 0306 caps for pretty wideband bypassing...
    looks reaaalllly nice.
     
  11. Joerg

    Joerg Guest

    Hello Joel,
    I usually go a bit lower than that in impedance. Four times Z is quite
    customary in inductive coupling though.

    Regards, Joerg
     
  12. John  Larkin

    John Larkin Guest


    If you have a multilayer board with close power and ground planes, any
    trashy caps, placed most anywhere, will be fine as bypasses.

    John
     
  13. John  Larkin

    John Larkin Guest


    OK, I'll try to draw it....


    A
    ----------------- ---------------------------------------
    | |__________
    ---------- |
    | |
    ---------------------------- ----------------------------
    B

    (pretty bad, huh?)


    So you put small ceramic caps horizontally at A and B, and fill in the
    middle, horizontal gap with vertically-oriented bigger stuff. Seems to
    work.

    I wonder if long interdigitated fingers would be good. The fingers
    themselves would be the highest-frequency couplers. Too hard to draw.
    If there's interest, I'll post pics of the guts of the PSPL tee, and
    the reverse-engineered schematic, to a.b.s.e. It's rated to 18 GHz or
    something.

    John
     
  14. Joerg

    Joerg Guest

    Hello John,
    When I open stuff like that I sometimes get upset that a $1000 device
    has next to nothing in there. Maybe a buck in parts, an aluminim case, a
    few screws and a couple of N-connectors. But when I have to make one I
    begin to appreciate it more, slaving an hour over it at the analyzer.

    Regards, Joerg
     
  15. Joel Kolstad

    Joel Kolstad Guest

    It got the point across just fine; thanks!
    I believe I can imagine what you're saying; interesting idea.
    I'd love to see it!

    ---Joel
     
  16. John Larkin

    John Larkin Guest

    OK, posted to a.b.s.e.

    John
     
  17. rex

    rex Guest

    Thanks for the nice hack.

    In pic 348x3 the teflon looks pretty solid.

    In 353x1 I guess we are looking at the blocking cap?
    Hard to figure how this relates to the other pic. There seems to be some
    slices to the left side of the pic. Wondering how the bias connects to
    the coax line. Is it shown (did I miss it).

    I think I almost ordered one of these or something similar on eBay
    recently. Maybe you outbid me.
     
  18. Why should it? They are both the same l/w ration, so I'd expect
    ESL to be about the same, just under a nH. If the capacitor is
    used for RF interstage coupling, it's part of a transmission
    line and its ESL is probably about the same as the length
    of transmission line it replaces.

    Jeroen Belleman
     
  19. Yes please, that'd be interesting. But could you put in a corner of
    a web site please? I have no a.b.s.e. here.

    Thanks,
    Jeroen Belleman
     
  20. Phil Hobbs

    Phil Hobbs Guest

    Depends for what. This was a satellite telecom system, with a truly
    nasty spurious response spec--5 Hz RMS in a 5-100 Hz BW around a 14 GHz
    carrier, plus equally unpleasant stuff further out. The 14 GHz had to
    be tunable in 1 MHz steps, and was derived from a 120X multiplier on a
    115 MHz reference, which was one of the boards I was designing. Made
    for interesting PLL work, especially since this was well before DDSes.

    Gave me a useful postgraduate course in shielding and bypassing.

    Cheers,

    Phil Hobbs
     
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day

-