References: <3DC1D056.firstname.lastname@example.org> <3DC1F17C.D3149AB3@SpamMeSenseless.us.ibm.com> <3DC21367.7060908@BOGUS.earthlink.net>
Subject: Re: Who thinks this?
X-Newsreader: Microsoft Outlook Express 5.50.4807.1700
Date: Fri, 01 Nov 2002 07:09:38 GMT
NNTP-Posting-Date: Fri, 01 Nov 2002 02:09:38 EST
Organization: Cox Communications
"Chris Carlen" wrote in message
> Phil Hobbs wrote:
> > I'll leave the LC transmission line questions to others, and just make
> > an observation about really fast interconnections: At very high speeds
> > (2 Gb/s or higher), transmission lines on circuit boards really do look
> > like distributed RC circuits, e.g. they slow down quadratically with
> > distance. The reason is basically copper losses and dielectric
> > absorption in common materials such as FR4 board. To get fast signals
> > to go any distance on a backplane, you have to do all sorts of ugly
> > preemphasis tricks to preserve bandwidth. In many instances, e.g. the
> > IBM Regatta servers, fast logic signals are demultiplexed down into
> > several lines each to cross the board or backplane, then multiplexed
> > back up into the original logic line. Ugly but necessary--this isn't
> > your dad's computer.
> > One reason computer companies (such as my employer) are looking at using
> > optical interconnections is that the bandwidth can be preserved over
> > much longer distances.
> Interesting. That indicates a case that is far from ideal, unlike what
> I am conjecturing about.
The PC board transmission line models, in a very short distance, the way
your transmission line probably acts over a long distance. In most
wire-based transmission schemes, there are distance limits that result from
various losses and distortions in the cable, no matter what the cable is. At
2.5Gb/s, a signal can only run through a 50 ohm PC board trace on FR4 for
about 20 inches before it has to be reclocked. The limit in a coaxial
transmission line may be longer, but the effects are essentially the same.
I think your ideal cable only appears to be ideal because your signal isn't
good enough to show you that it's not. If you have a good TDR, you can
easily see the losses in even short sections of cable, as well as
connectors, terminations, and so on.
> I'd like to master the totally general simulation of transmission line
> wave propagation, using the transfer function the whole way through.
> But at this point I only know how to do steady state AC, and step/pulse
> propagation with real reflection coefficients and Zo.
There are myriad texts in this field, many of which date from the late 1910s
to early 1930s. Some of the first work in the field was performed by William
Thomson (Lord Kelvin) and George Stokes in the late 1850s and early 1860s.
-- Mike --