Re: Prototyping?

John Larkin jjlarkin@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx posted to
sci.electronics.design:

On Fri, 02 Nov 2007 10:12:56 -0700, Joerg
<notthisjoergsch@xxxxxxxxxxxxxxxxxxxxx> wrote:

John Larkin wrote:
On Thu, 1 Nov 2007 17:29:52 -0700, "Joel Koltner"
<JKolstad71HatesSpam@xxxxxxxxx> wrote:

"Joerg" <notthisjoergsch@xxxxxxxxxxxxxxxxxxxxx> wrote in message
news:kGtWi.2430$yV6.1485@xxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Agreed. I breadboard only when trying to use parts in really
unorthodox ways (happens a lot...) or when I need a one-off to
control something and it doesn't have to be pretty. Beats the
usual 1-2 week wait for fab, stuffing and all the Fedex in
between.
OK... but say for something as "simple" as a UHF bandpass filter
(say a standard ham band one... 420-450MHz, assume you've decided
you need a 5th order Chebyshev filter to obtain the skirts you
want), do you (or John) expect you can design and layout a
working PCB without either...

1) Performing simulation using one of the high-end CAD tools like
Genesys, ADS, or Microwave Office, which can use very good models
of the capacitors and inductors you're wanting to use as well as
accounting for most parasitic effects of pads and trace width
variation.

LTSpice should be enough at 450 MHz. Or even just yank a prototype
filter from Williams' book and scale it. With surface-mount parts,
450 MHz isn't really different from audio.


Even 2.45GHz is done with discretes these days. What surprised me
last week (UHF filter design) was that nicely toleranced caps where
only $0.01 and inductors $0.05. That was almost as good as spotting
a favorite microbrew on sale.

We mostly do untuned wideband stuff, DC to as high as 4 GHz, which
is fine with surfmount parts on FR4. But if you want to do
narrowband filters, at some point you have to tune them, or go to
saws or coaxial ceramic resonators or something, since the discrete
tolerances will getcha.




2) Fully intending to perform a fair amount of tweaking
(generally of capacitor values, given the frequencies involved)
once you actually build the board.

Why tweak a prototype kluge, then move it to the real PCB, and
tweak it again? If you expect to have to tweak a few cap values to
get it to work, why waste time doing it twice? Besides, complex
filters are almost impossible to tweak experimentally, unless
you're willing to tolerate truly rotten passband response.

???

I certainly can't do so myself, and if someone else can I'd love
to learn
their tricks! The fact that I've never read of such a method in
amateur publication (books, magazines, etc.) suggests to me that
if probably isn't doable... and the commercial guys just go with
method #1 up there.

I don't think it will offend anybody to point out that amateurs
are amateurs.


Au contraire. I found that there still is a decent population among
the ham radio community who can design a nice GHz range filter and
actually make it work. Maybe only a few percent but IME the
percentage of folks coming out of university who could do that is
nearly zero.

Our problem isn't to make one filter work. It's to make an entire
board work: bus or comm interface, power supplies, uP, firmware,
fpgas, dacs, adcs, filters, output amps... all in maybe 90 days. We
don't have the time to breadboard, or even play with eval boards,
except for small, rare situations. We certainly don't have time to
prototype any serious subsystem. So we lay out the real thing, fully
documented and formally released, and let Production build one or
two for us to test. That saves scads of time and hassle - they
solder better and cheaper than engineers - and if you're careful,
the first one is the real thing.

Intel has, or at least had, a similar philosophy: new chip ideas
were fabbed on the production lines, not in some backroom lab.

John

That reminds me of when i was working for Hughes, both the Carlsbad
and the Santa Barbara research teams could reliably turn out "first
of", or "one of a kind", or "unique" high performance devices. But
do not ever try to get production quantities, they could not do it,
and they were no longer interested. If you did the yields were
spectacularly poor (approaching less than one good part per wafer for
IC' and imagers). It was quite the joke internally.

.