morrison@nucsrl.UUCP (Vance Morrison) (02/08/88)
Hello, I have a question for the ethernet experts. I have heard some people mention that the restrictions on segment lengths and transceiver placement (on the black lines), and I really do not see the rational behind it. Oh, they mumble something about reflections off the transceivers and the couplers, but this to be is a simple minded answer. I happen to be an EE and my understanding of transmission lines makes the answer to the cabling restrictions far from obvious. First of all, we are not sending a single frequency down our transmission line, for ethernet, the signals can be anywhere in the range of 5-10MHZ (because of the manchester encoding), depending on the bit pattern being transmitted. Now certainly we will have reflections off couplers, transceivers and any other discontinuity, but whether these reflections add or cancel depends on the bit pattern, and thus is unknown. The ONLY way to insure that these reflections have no adverse effect is to insure that they are small (not by hoping they cancel). Thus it seems to me that there is no benefit in the cabling restrictions. If anyone knows that rational, please let me know, because the restrictions do not make sense to me. Vance Morrison morrison@accuvax.nwu.edu morrison@nuacc.bitnet morrison@northwestern.arpa PS. Along the same lines could someone please tell me the reasoning behind the minimum transceiver distance.
kwe@bu-cs.BU.EDU (kwe@bu-it.bu.edu (Kent W. England)) (02/11/88)
In article <3880004@nucsrl.UUCP> morrison@nucsrl.UUCP (Vance Morrison) writes: > > Oh, they mumble something about reflections off the transceivers >and the couplers, but this to be is a simple minded answer. I happen >to be an EE and my understanding of transmission lines makes the answer >to the cabling restrictions far from obvious. > Let's see if I can help answer this without mumbling. I am an EE and I studied transmission lines and antenna theory, but I don't want to try to recall the theory and fill the page with differential equations. I don't think that's what you want anyway. First, as engineering students we studied analog signals or modulated analog signals and we didn't worry about the instantaneous time histories of these spectra. Ethernet uses baseband (square wave) signals with spectra that vary instantaneously, but for the purposes of understanding the reflections and interference patterns we can use average spectra or instantaneous spectra to good effect. By that I mean take a sequence of a few bits in a packet, figure the spectrum, and see what happens on a given transmission line (the Ethernet cable). In general, there will be a 1st, 3rd, and 5th harmonic for a uniform square wave. These will set up varying standing waves on the cable. If you are located at an interference point (which, of course, varies in location instantaneously) your spectrum will be corrupted and your received waveform will not be "square" and, according to the rules for decoding Manchester signals, you might not find the right spot for sampling a stable level. In addition, the rules for detecting collisions depend on average (I don't know how the average is computed) power levels at your transceiver. If the instantaneous power levels received average out a "quantum" higher than expected due to the instantaneous interference received over the averaging interval, a collision will be declared and the packet aborted. Now this instantaneous behavior must hold true on average for the general rules about cable lengths and transceiver spacings to be valid. You only have to corrupt a bit or two to lose the packet, so if the instantaneous interference point location wanders over your location on the cable, you will experience bit errors. Perhaps a DEC engineer can explain how their lab tests prove out their recommendations on cabling. Otherwise, you can try systematically violating the rules and then we can put a TDR and a LAN analyzer on your net and see what's what. :-) This explanation wouldn't help you do RF R&D, but it's only intended to be a rough model (which is still allowable in the spirit of scientific inquiry). Is it helpful without being too inaccurate? Kent England
ron@topaz.rutgers.edu (Ron Natalie) (02/12/88)
> I have a question for the ethernet experts. I have heard some > people mention that the restrictions on segment lengths and transceiver > placement (on the black lines), and I really do not see the rational > behind it. The restriction on Ethernet transcievers (IEEE 802.3 MAU's) are that they be located 2.5 meters from each other (minimum). The black marks on the cable are located 2.5 meters to be used as a guide. This is especially handy when the cable is in the cieling and you can't see 2.5 meters in each direction on the cable, so always putting them on the black marks is a safe way of handling it. The 2.5 meter marks have NOR relationship to the end of the cables. The standard (IEEE 802.3, the only one I have handy) specifically says that the mark spacing is discontinuous at the connectors in the cable. The main reason the marks are there is that the standard for the cable properties lists insulation with these markings. Note another useful marking is cable "serialization." These are numbers that are marked every so often down the cable with constant spacing. Knowing the number at the end of a segment will allow you to compute the distance to a number further down the cable. This makes TDR work possible when you have random cable in the cieling. > I have a question for the ethernet experts. I have heard some > people mention that the restrictions on segment lengths and transceiver > placement (on the black lines), and I really do not see the rational > behind it. If you look at a frequecy domain plot of it, you'll find it at 10MHz (plus harmonics, square waves are full of 'em). The frequency is always 10 MHz, the Manchester encoding changes the phase of the square waves by 180 degrees. If we didn't use manchester encoding, then the frequency could become 5MHz often, and even DC for a while, which would be bad. > Now certainly we will have reflections off couplers, transceivers > and any other discontinuity, but whether these reflections add or cancel > depends on the bit pattern, and thus is unknown. The ONLY way to > insure that these reflections have no adverse effect is to insure that > they are small (not by hoping they cancel). Thus it seems to me that > there is no benefit in the cabling restrictions. The major place reflections occur is at impedence mismatches. The main reason for impedence mismatch on Ethernet is cable joining, not the MAU's themselve (unless they are inline, which effectively adds two connectors to the system). This is why there are some guidelines (restrictions?) for wire lengths in the system. They go, in decreasing order of preference. 1. Make the Ethernet out of one piece of wire (no discontinuities). 2. Make the Ethernet out of pieces of wire from the same manufacturer and lot (i.e, from the same spool). This has no cable to cable impedence changes, any problems will be those of connectors. 3. If you must mix random cable pieces, use lengths that will cause the reflections not to add in phase. The standard points out that if you use half-wavelenths of 5MHz for all your cable pieces, you won't have a problem before using up your 500m of cable. These lengths are the familiar ones (23.4, 70.2, 117, ... meter) that your mama told you about. 4. There is actually a measurement test that can be done to verify the allowable configuration with trully random lengths and parts. I don't have my IEEE 802.3 handy (it's at home). But most people don't have the equipment to do that anyway. Generally, I usually refer people to the DEC "Networks and Communications Buyers Guide" which explains many of the Ethernet design criteria. They leave out a lot, but the are conservative, so if you put together a network based on their specs it will work. -Ron