From: John Miles (jmiles@pop.removethistomailme.net)     Message 4 in thread
Subject: Re: Wireless ethernet??
Newsgroups: rec.radio.amateur.digital.misc
Date: 1999/06/10                           View complete thread (4 articles)

Mikael Östh wrote:
>
> How about building your own transmitter and reciever for the ethernet
> protocoll!!??
>
> Is that possible ?
> suggestions??
>
> Regards
> Mike
> osth.teknik@telia.com

This message was sent via a homebrew 10-megabit Ethernet link that's
been running for about 2 weeks now, so yes, it is possible. :)

One approach, by Fredrik Oberg, is described at
http://w1.911.telia.com/~u91105482/indexeng.html.  Fredrik's circuit has
a few problems as detailed, but the last I heard from him, he was
achieving at least some limited success at 10 megabits.

With any luck I'll have a good description of my design up on
http://www.qsl.net/ke5fx within a couple of weeks, complete with
schematics and .JPGs.  It'll depend on how much time I have over the
next couple of weekends to take everything back apart and draw the
schematics.  Ralph Stirling, KC3F, has graciously volunteered to
translate my back-of-the-napkin schematics into a legible CAD format,
but at the moment the bottleneck is entirely on my end.  I will drop an
announcement here when the web page is functional.

In the meantime, some general hints:

- The N6GN 2-megabit link at
http://www.tapr.org/~n6gn/uwavelink/uwv.html (see also the 1993 ARRL
Handbook) is the usual starting point for those who are interested in
10-megabit work.  Substitution of the faster MC13155 for the MC13055
used in Glenn's design is the obvious first step on the road to
10-megabit compatibility.  There have been various efforts to
"underclock" 10-megabit Ethernet cards down to 2 megabits for direct
compatibility with Glenn's circuit, but I am not sure if people are
having any luck with this approach.

- The 10 GHz Gunnplexers sold at http://www.shfmicro.com are great for
this purpose.  They have built-in varactors with good modulation
characteristics for wideband FSK (as opposed to the technique of
voltage-modulating the Gunn diode directly in the original N6GN
design).  The 10-milliwatt Gunn diodes can be upgraded to 100 milliwatts
if necessary.

- Watch out for DC level-shifting issues!  It is NOT a good idea to feed
an FSK signal at +/- 2V or so to a +8V-biased varactor through a large
coupling capacitor. The coupling time constant will mangle some or all
of your packet traffic.  An amp such as the OPA603 can be used to apply
the desired DC offset to your varactor drive signal without any AC
coupling at all. I spent way too much time ignoring this type of
problem.

- Further to that, in my experience it's best to treat the link's entire
signal chain as an analog, rather than a digital one.  Your Ethernet
hardware (NIC, hub, whatever) has a very broad tolerance for out-of-spec
signalling levels, waveshapes, and the like.  It's unnecessary to use
comparators, ECL line drivers, and the like in your transceiver units.
(In fact, use of two-state comparators adds an extra dimension of
polarity sensitivity to the problem.)

- In any dual-Gunnplexer design, you will need to use the AUI
(Attachment Unit Interface) port on your network hardware.  The
immediate temptation is to use 10BaseT directly, but the local echo
generated by the Gunnplexers will be interpreted as a collision on every
outgoing packet under 10BaseT.

- Contrary to often-stated dogma from the Ethernet gurus, collision
detection is not worth losing sleep over in the real world.  If
separation of two busy collision domains is required, simply use bridges
or switching hubs to calm things down.  A two-station point-to-point
link will end up looking like an AUI crossover cable, which is an
illegitimate but perfectly-functional hack.  The collision pairs on the
AUI ports are left unconnected.

More later, hopefully.

-- jm KE5FX

------------------------------------------------------
Note: My E-mail address has been altered to avoid spam
------------------------------------------------------