Re: Feliz Haridad
Ron Miller:
At 03:40 PM 17-12-99 -0800, Ron Miller wrote:
>Hi Pat and Rich
>
>Let me help ease some of the confustion on bandwidth requirements:
>[...]
>5. Bottom line is, specify the filter by: shape(bessel), number of poles(probably 1, 2 or 3)
>and by the attenuation specified at the frequency. [...]
we had a small bit of difficulty with this waaaaay back in 10BaseT days, at a
time when the proposed version was to include an external filter module
in the transmit path.
at that time the task force was asked to define many electrical characteristics
of a link, and with that, recommend minimal characteristics of reflection, insertion
loss, transition band slope, attenuation, and so on, for this module. after a bit
of work some of us found that most characteristics were relatively easy to
realize with off-the-shelf components, but that desired return loss was not
as easy.
we concluded that for most characteristics you could use either 1 pole
(which did not offer enough transition band slope) or 5 poles, but nothing
in between offered sufficiently flat impedance variation over frequency , from
a 2-port perspective. to be fair, we did not characterize things well enough
to explain exactly why this was true (though if memory serves it had a lot
to do with manufacturing variability and design margin). it's my perception
the committee were mostly interested in arriving at an acceptable answer,
and this was sufficient. module vendors wound-up offering something
like 8 or 9 poles, and all of that was supplanted by the filterless version.
finally, so-called Bessel-Thompson filters offer a nice tradeoff between
transition band slope and group delay distortion, hence their relatively wide
acceptance in datacom. i'm not certain i know what a "Bessel" filter is.
--
J M Wincn, Staff Engineer
Cielo Communications, Inc.
325 Interlocken Pkwy, Bldg A
Broomfield, CO 80021-3497
Voice: 303-464-2264
Fax: 303-460-6370