Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!uakari.primate.wisc.edu!aplcen!haven!decuac!bacchus.pa.dec.com!decwrl!sgi!rpw3@rigden.wpd.sgi.com From: rpw3@rigden.wpd.sgi.com (Rob Warnock) Newsgroups: comp.dcom.lans Subject: Re: 50-75 ohm impedance matching Message-ID: <64296@sgi.sgi.com> Date: 17 Jul 90 05:00:33 GMT References: <1990Jul13.210239.17750@ux1.cso.uiuc.edu> <8601@uhccux.uhcc.Hawaii.Edu> <28221@cc.usu.edu> <8611@uhccux.uhcc.Hawaii.Edu> Sender: rpw3@rigden.wpd.sgi.com Reply-To: rpw3@sgi.com (Rob Warnock) Organization: Silicon Graphics, Inc., Mountain View, CA Lines: 57 This is NOT going to work, guys... In article <8611@uhccux.uhcc.Hawaii.Edu> querubin@uhccux.UUCP (Antonio Querubin) writes: +--------------- | In article <28221@cc.usu.edu> BOBW@cc.usu.edu (BOB WOOD WA7MXZ) writes: | >Look at your autotransformer design again. Not only does an autotransformer | >pass DC but it also shorts it out, directly to ground. The larger the wire | >used in the transformer to improve the Q the lower the D.C. resistence to | >ground. In short, the matching transformer plan isn't going to work. | Yes, you're quite right. But considering that the DC signal is a relatively | short pulse on the order of milliseconds at the most the autotransformer may | work anyway. The trick is to make the thing have wide enough bandpass to | accept both the low frequency pulse and the high frequency component. +--------------- Nope, the D.C. signal is *not* necessarily "short". Consider a long stream of back-to-back long packets, and note that the gaps between the packets do not allow time for the D.C. balance to "relax". There is no limit to the maximum length of the D.C. "pulse" as seen by the autotransformer. Eventually you will get enough current in the autotransformer that the voltage will (1) not get high enough *during* packets, and (2) start kicking the "wrong way" *between* packets, which will mess everything up. By putting a (carefully chosen) capacitor between the bottom of the auto- transformer and cable shield, you *may* be able to avoid that problem. However, see below... +--------------- | >This is going to be your only game plan. Use only 2 nodes on the cable and | >use good 75 ohm 1% terminating resistors. +--------------- Nope, this doesn't cut it either. Ethernet transceivers are *current* drivers. The collision-detection logic measures *voltage*, which is the sum of the currents into the cable times the impedance of the cable. Raising the cable impedance (and the terminating impedance) by 50% (50 -> 75 ohm) will raise the voltage by 50%, and you'll constantly be on (or *over*) the ragged edge of a (false) collision detect. Notice that the Ethernet spec says that transceivers must detect collisions within two bits times, so no matter what D.C. termination you put on the cable, as little as 40 meters of 75 ohm cable will cause constant "collisions". The only thing you can do is use the autotransformers, *with* the blocking capacitor I mentioned above in the ground leg, and 50-ohm cable and 50-ohm terminators at the ends. All transceivers must be on the 50-ohm sections. And even then, no guarantees it's going to work... Good luck! -Rob ----- Rob Warnock, MS-9U/510 rpw3@sgi.com rpw3@pei.com Silicon Graphics, Inc. (415)335-1673 Protocol Engines, Inc. 2011 N. Shoreline Blvd. Mountain View, CA 94039-7311