1 ghz lowpass filter

[Michele Ancis]

Hi, Michel,

What I'm making is a programmable arbitrary waveform generator to
modulate this laser. It works by summing a bunch of individually
programmable fast impulses, spaced 250 ps apart. The filter smooths
them into, ideally, a nice continuous waveshape, and we use waveform
feedback, from further up in the optical system, to tune the impulse
train to get the actual optical pulse shape we need. So what I need is
actually a filter with a Gaussian time-domain (impulse) response,
which is about what a Bessel does; but there's no hard
frequency-domain performance requirement, and the filter response is
convoluted with a lot of other elements in the chain anyhow. So I
figured I'd start with a Bessel, and tweak it (x-acto knife? crazy
glue dielectric blocks?) until the overall system response is nice.

Along the way, it ocurred to me that a series of Gaussian impulses is
a nice way to do curve fitting, as opposed to polynomials or Fouriers
or whatever. If the Gaussian impulses are spaced so as to overlap at
about their half-amplitude points, they can be tweaked nicely to
follow most any reasonably smooth function.

John

Hi John,

ok now I start to see what you intend to do with your filter...Just
keep in mind that the one proposed by Frank, at least to my
simulations, does NOT have a Bessel response. As you know, the
frequency response must be characterized through both amplitude and
phase. Or, in our case, amplitude and group delay. What I'm saying is
that, since the group delay is not constant, you can't expect to have
a simmetrical gaussian pulse response from Frank's filter (unless my
simulations are completely wrong :-( ). I've simulated my lumped
element - ultra-ideal-with-theoretical-values - filter, and the one
proposed by Frank (wich has ideal microstrips). The result is a nice
"gaussian" pulse response with mine, and something that resembles a
gaussian shape but is not simmetrical and has ripple with Frank's one.
I don't know what you're looking for, it is just to warn you about
what you're about to get, according to my understanding.

However, the ripple may well be generated by the time domain model of
the microstrips, therefore I wouldn't trust it that much.

On the other hand, I'd rather be sure that, without equal group delay,
you won't get simmetrical pulses, and your control over the waveform
will possibly get more difficult. As you'll have noticed, also the
"bandwidth" of the filter is related to the width of the pulses
(responses in TD), therefore to the capabilities of your waveform
generator to create "arbitrarily" fast functions...if you're slow
enough, maybe you can obtain nice simmetric pulses, but don't expect
too much if you want to drive your filter "harder". I mean, there's a
portion of the frequency spectrum where the response of Frank's filter
has nearly constant group delay (after all, this happens with all
kinds of LP, more or less): if the "relevant" frequency content of
your synthesized waveform is in that portion, everything should be
fine. But if you intend to change its amplitude too fast, you will see
distortion. This is, at least, what I can understand...

Please keep me informed as you go ahead,

Michele (not Michel, which is french!)

 

[John Larkin]

Hi John:

The "lumped element" approach is not out of the question.
Have a look at the Coilcraft site www.coilcraft.com They have several
ranges of SM parts with SRF's well above 1GHz. For example the
air spring types offer 1.65nH-12.5nH with SRF's all above 4.6GHz.
There are a number of specialist HF capacitor manufacturers that would
also be suitable, but may be overkill. I've done a filter with 500MHz
passband that was still >50dB down at >2GHz just using ordinary
0805 SMT caps. I was able to "breadboard" the filter using the
Wainwright
SM PCB kit (although a few bits of PCB and a scalpel would be almost
as good in this case) with pretty good match to the final board.

Regards
Ian

Ian,

Yes, this thing is just barely doable with surfmount parts. The
inductors are no problem - I can get 0603's down to 1 nH, and I only
need to go down to 5 or so - but the caps get smallish, 0.7 pF in one
design. Those small caps only come in hefty value jumps, which might
be awkward, and a trimmer cap might well have too much L. Maybe a
hybrid, with 0603 Ls and the really tiny caps being copper patches?

Trimming copper is like cutting hair: you can cut it shorter, but you
can't cut it longer. Of course, in addition to the filter, we have
about 68 zillion more, equally nasty problems to deal with here.

John

 

[John Larkin]

OK John, this time I'll forgive you ;-))

oh...that's very simple...throw the numbers in a sim program!!


I have some considerations to do about this:

1) All distributed element filters (i.e. stubs AND stepped impedance)
exhibit some repetition of the attenuation pattern. The stub based
filters are periodical because the lines are generally of the same
lenghts. In the stepped impedance filter, the lines' lenghts are not
in a "nice" ratio (i.e. a natural number), therefore the pattern will
repeat but NOT with periodicity. But SURE it will repeat. It will pass
some other frequency ahead in the spectrum.

2) You keep talking of amplitude response, but please note that the
phase response, or the group delay, is also important in your
application. I would say it IS the most fundamental thing. If you're
only looking for nice attenuation, why bother with Bessel? You can
quickly realize a lumped element Butterworth or Chebycheff with far
better selectivity AND you can do it with SMA components!

3) Concerning lumped Vs. distributed, remember however that the SMA
are real components either. At some point, caps will act like
inductors and vice versa...but I guess you knew it.


The reference I gave you some posts ago is something similar, but far,
far more! It is a very good book.


Ok, I'll wait with patience!

Michele

Oh, Amazon just presented me with "Microstrip Filters for Microwave
Applications" by Hong and Lancaster (no, not *that* Lancaster.) They
clarify somewhat the spurious stopband issue: stub filters are always
nasty, but stepped filters *may* have monotonic stopband attenuation
if conditions are right. The concept is to approximate a
lumped-element filter as closely as possible, which means using very
skinny, very short lines for the Ls and wide patches over thin hi-K
dielectrics for the Cs, so that all geometries are small compared to
signal wavelengths and the filter is, well, lumpy. Presumably if you
push this far enough, losses will take care of any very high-order
responses. They have a nice dramatic graph of LC filter + bad stepped
filter + good stepped filter.

This book also has the long-sought closed-form expression for the
capacitance between parallel microstrips. But it's a page full of
equations-inside-equations, hyperbolics and elliptics all over the
joint; I may assign my scut bunny to try to code it up.

John

 

[Scott Newell]

inductors are no problem - I can get 0603's down to 1 nH, and I only

need to go down to 5 or so - but the caps get smallish, 0.7 pF in one
design. Those small caps only come in hefty value jumps, which might

Have you looked into the Vishay silicon caps? I saw 'em
(HPC0402 series?) in the latest Mouser...looks like they've
got 0.6, 0.7, 0.8, and 0.9pF in stock. The datasheet shows
0.1pF to 1pF in 0.1pF steps, with a loose 0.05pF (best grade)
tolerance.


newell

 

[John Larkin]

Does this book cover any lumped designs based on any copper shapes
other than rectangle elements? I have a strong hunch that rectangles
are used because of the ease of laying out and describing, not because
they make the optimum elements from a performance standpoint.

Oh yeah. They have filled circles, rings, triangles, serpentines,
ladders, trumpets, and some really cool looking fractal rings. He
talks about high-temp superconductive filters, too. "Only" $62, not
bad for this sort of thing.

John

 

[John Popelish]

Oh yeah. They have filled circles, rings, triangles, serpentines,
ladders, trumpets, and some really cool looking fractal rings. He
talks about high-temp superconductive filters, too. "Only" $62, not
bad for this sort of thing.

Cool!