Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!usc!rutgers!bellcore!jupiter!karn
From: karn@jupiter..bellcore.com (Phil R. Karn)
Newsgroups: comp.dcom.lans
Subject: Re: Ethernet over twisted pair: a summary of the "whys"
Message-ID: <18510@bellcore.bellcore.com>
Date: 6 Dec 89 23:08:59 GMT
References: <4686@blake.acs.washington.edu> <2230039@hprnd.HP.COM>
Sender: news@bellcore.bellcore.com
Reply-To: karn@jupiter.bellcore.com (Phil R. Karn)
Organization: Bell Communications Research, Inc
Lines: 49


On the subject of Ethernet and RF interference, I can say with some
authority that coax is *not* a guarantee against problems.

I have a thin Ethernet at home with three machines (Sun, XT and 386 PC
clones). When I replaced my PCs' old decrepit 3Com 3C-501 Ethernet
controllers with newer TRW PC-2000 boards, the level of RF noise heard
in my HF (< 30 MHz) amateur radio transceiver went up significantly. It
seems that as competition heated up in the PC Ethernet market, the
manufacturers started cutting corners on their RF filtering.

The root problem is the requirement that Ethernet coax shields be
isolated from the host computer ground. Ethernet controllers with
built-in thinnet transceivers do this with an insulated BNC connector on
the back panel. Unfortunately, the isolation between the transceiver and
the rest of the unit is not very good at RF frequencies (probably due to
capacitive coupling across the isolation transformers, plus other stray
board capacitances). The end result is that the coax acts as a nice
long-wire antenna that effectively conducts RF noise out of the computer
and radiates it. Note that the coax shield is ineffective here because
the noise is a common-mode signal; radiation is from the outside of the
shield.

The very early 3C501 cards had reasonably effective pi-network low pass
filters on their BNC connectors that consisted of ferrite rings plus a
couple of disk capacitors from each side to the metal mounting bracket.
Later versions of the 3C501, plus just about every other manufacturers'
cards, started scrimping on the filters. Some current cards have no
filtering on the connector at all, but instead have a single disk
capacitor/spark gap on the PC board between the transceiver and PC
grounds. This probably *enhances* RF leakage because of the "hot"
internal grounds found in most computers. (Many digital engineers have
apparently never heard of "ground loops".)

I finally got my problem under control by using external transceivers.
Ethernet transceiver cables are well shielded, and more importantly
their shields connect directly to the external metal casing of the PC
and to the metal chassis of the transceiver.  Also, the capacitance of
the cable apparently does a good job at attenuating conducted noise
between the signal pairs and the cable shield. It's still not perfect,
but about the only better thing I could try would be fiber.

By the way, for a good time try monitoring an Ethernet with a shortwave
receiver tuned to 10 MHz. File transfers sound remarkably like the
infamous Russian Woodpecker (a Soviet over-the-horizon radar that is
much reviled by radio amateurs and other users of the HF spectrum
because of the widespread interference it causes.)

Phil