From: John Tserkezis 
Reply-To: Fidonet AVtech Echo 

Jasen Betts wrote:

>  JT>   As I understand it, it's still 50 ohm.  The terminators make the
>  JT> ends look like an infinite length cable so there are no reflections.
> 
> It is 50 ohm cable but the card is driving two stretches of cable in connected
> to it in parallel (by the Tee connector), so it sees a 25 ohm load.

  Ok, that sounds logical too.  But they never taught us that in transmission 
line theory.  The bastards only concentrate on ideal models, never real life 
examples.

> IIRC Signals travel in coax at arround 200 Mm/s (2/3 lightspeed)
> so a 50 ns pulse (10Mbps ethernet) is about 10m long

  Yes, about 60% the speed of light on your joe average cable.

> That short.

  Long enough to see on a cro.  In fact, we used that to find cable breaks in 
the office coax layout.  The original networks was put in by six different old 
employees where none of them were competent at crimping their fingers in a vice 
let alone properly terminate an ethernet coax cable.

  It was an el-cheapo time domain reflectometer, one signal generator, one cro. 
  Leave the other end underminated, (or shorted) and look at the reflections. 
We estimated distance going on the time of the reflection.

  Worked quite well.  Saved us from reterminating the whole network.

  I only wish I could have a portable version for use at client's offices, 
where after spending an hour looking for cables faults, you finally find it at 
a badly terminated BNC connector under some guy's desk, where he THEN promply 
says "yeah, it does that every time I kick it".

> Hmm, maybe I should repeat that experiment with a shorter cable.

  Naw, the longer the better.  If you have t-pieces inbetween, you can 
experiment and progressivly remove each one and look at how it effects the 
reflections.

> conclusions
> a 25 ohm load is needed
> a 10m cable is long enough for reflections (or its impedance when unterminated)
> to be a problem at 10Mbps
> a 1m cable is isn't.

  Interesting.  I had never tried it, we just went on typcial recommendations, 
and when something didn't work, it was due to some fault, not one of our 
"experiments".

>  JT>   In theory, cable length should not be an issue (aside perhaps from
>  JT> degraded signal strength by the time you get to the end of that 300
>  JT> metre stretch), although I've seen some cards have a _minimum_ cable
>  JT> specification (about two
>  JT> metres if I recall).
> 
> hmm... I wonder what that minimum does.

  I suspect it's to ensure it works within the card's dynamic range.  Too 
strong a signal, like any reciever, would swamp the input stage and reduce 
quality of signal.

>  JT>   Again in theory, if you had a long enough cable length on one (or
>  JT> both) ends, the system should still work, because there are no
>  JT> reflections, thus no corruption of signal. What you would need is a
>  JT> cable length long enough at the ends that the inevitable reflections
>  JT> are degraded enough by the time they come back, that they offer little
>  JT> or no effect on the closer valid signals. --
> 
> or that they take long enough to come back that the original has already
> reached it's destination :)

  Er, that's not one of those 'if a tree falls in a forest' or the 'sound of 
one hand clapping' question is it?

> I wonder how much loss you get in 100m of that cable at 10Mhz.

  Easily calculated.  Cable specs state loss at specified freqency use, you 
just add it up for your length of cable.

  T-pieces add a bit of loss too, but largely irrelavant over your typical coax 
network.

-- 
       -o)
       /\\    Message void if penguin violated
      _\_V    Don't mess with the penguin

Linux Registered User # 302622                         http://counter.li.org>
Fido: 3:712/610  BBS/FAX: +61-2-9716-8310  Internet: jt{at}techniciansyndrome.org
--- ifmail v.2.15