On Wed, 26 Oct 2005 13:36:41 +0100, Paul Burke wroth:
A hybrid at 10 MHz might be beyond the capabilities of ordinary ethernet transformers. If I had access to a network analyzer, I'd wire one up first just to see if the forward/reverse isolation was good enough.
If it wasn't, I'm sure something could be put together that would do the job. Maybe a couple of MiniCircuit boxes?
First of all: please do not multipost. I see that this message appears in comp.arch.embedded, but without searching for all copies of your message, it is impossible to tell if someone has answered your question correctly or not. It's a big waste of everyone's time. Crosspost if you wish, but avoid multiposting.
Unlike the other two responses that you've received so far, which are not particularly negative, Fast Ethernet on only two wires should definitely not work. In 100Base-TX, the transmitters are constantly transmitting a signal, even if your application does not produce any frames. This idle state transmission is done so that the receiving PHY does not lose RX clock snychronization. This differs from 10Base-T where data is manchester encoded and thus incorporates a clock with the data. Although it's unclear to me what you mean by a hybrid Ethernet transformer configuration, I would expect that any simple scheme which reduces you down to one wire pair would result in a never-ending collision situation. In all probability, the PHYs wouldn't even link up.
As far as I know, 100Mbps Ethernet has never existed on coax or any 2 wire scheme. Briefly looking at 802.3-2002, it appears that it only exists on 2 pairs of Category 5 copper (100Base-TX), 4 pairs of Category 3 copper (100Base-T4), or 2 optical fibers (100Base-FX).
Reading it a second time doesn't waste time to speak of. People quickly skip over messages if they discover that they've already read the message. The problem with multiposting is that you can have one person answer your question in one group, and I could not know about that and also compose an answer to your question in this group. Most of my Usenet postings, I take very seriously and spend 30 minutes or more composing (thinking about, writing code, rewording to avoid flame wars, etc). If someone has already suitably answered your question, me writing any message at all would be a complete waste of my time.
Multiposting also takes double the storage and distribution bandwidth compared to a crossposted message. This is of no concern to most people, but it is inefficient.
I'm not familiar with telephones. It sounds like what you are describing is what I'm going to call an "analog type of XOR". Adding the local transmission to the reception doesn't destroy the reception data because the local transmission is known and can be subtracted prior to playback on the earpiece.
This comment was intended to imply that I think John Larkin's statement was incorrect with regard to 100Mbps Ethernet. Passive coax converters may exist for 10BaseT, but I doubt they exist with 100BaseTX. In all actuality, I don't think there is such a thing as passive coax converters for 10BaseT either. I'm guessing that, he is remembering AUI to Coax and AUI to UTP converters, which are whole new active PHYs.
Why should it work? As far as I can tell, you would need to violate the speed of light to make it work. As I stated, the 100Base-TX PHY is constantly transmitting data, weather your application is idle or actively transmitting a message. On Cat 5. twisted pair, the signals propagate at
0.71C or 4.7ns/m. This means that at 100Mbps, each "average" bit only occupies 2.1 meters of cable. To over simplify, as two PHYs are both transmitting on either ends of a single pair cable, each signal could undergo several sum operations (mid-cable collisions) before reaching the other PHY. In order for the receiving PHY to recover the original state of the transmitted signal, it must XOR the received signal with all the signals that it collided with along the way. I don't care how cool a hybrid transformer configuration may be, it will not have a memory of previous signals transmitted and know how many of them to XOR with the incoming signal.
Even if your cable is less than 2.1 meters, the transformer XOR operation shouldn't work. For part of the bit, the signal would be recovered correctly, but the other part would be corrupted 50% of the time. This would introduce high frequency glitches into the data stream, which the PHY probably can't handle.
Seriously, I have a good excuse. I am quite young. It was only last year that I got my EE degree. I'm not a teen, so I don't know everything. I'm also not older than everything, so I don't know everything. These two things, in addition to an occasional desire to play games, leaves an opportunity for gaps in my knowledge.
Actually, it doesn't make sense, because it's wrong.
Signals going in different directions do not "collide" in the middle of a cable, except from the point of view of an evesdropper who has tapped the middle of the cable. Otherwise they pass harmlessly by each other.
But there is a potential problem: unless you very carefully terminate the line impedance at each end, the signal you are sending may bounce off the other guy's receiver and come back to you, corrupting what he is trying to send to you. The higher the frequency of signalling, the harder it is to get this termination perfect enough. Note by the way that terminating a line you are driving means you drive it through a series resistance equal to the line impedance, so right off the bat only half of the (AC) voltage of the driver's output shows up on the line.
If you could use perfect termination on each end of the line, you should be able to simply subtract what you are driving right now, to find out what he is saying to you... or are there other complications?
I think practially speaking it probably won't work reliably, and you are on the right track as to why, it's just the collision occurs due to impedance mismatch causing reflection - this echo is travelling in the same direction as the other guy's signal and so does "collide" with it.
And to be fair it's not just the ends of the cable - any change in impedance anywhere in the line can cause a reflection - bad connectors, some sort of splice, kink where the center conductor cuts through some of the dielectric and gets closer to the shield, etc.
Scarcely. Don't read it if you don't want to. But thanks for your response.
Like in telephones. You can talk both ways at the same time, over the same pair, but the transformer configuration cancels the locally transmitted signal from being received at the sending end. They actually have to unbalance the signal a bit so you can hear what you are saying yourself.
I know it does. But why shouldn't the hybrid configuration work?
On Fri, 28 Oct 2005 08:03:48 +0100, Paul Burke wroth:
Howard's been helping me on the software side of ethernet stuff, but it doesn't look like he's a hard-core hardware hacker. 8-)
It should. In theory, that is. In practice though, the hybrid network would have to compensate for the line impedance to a fairly high drgree of accuracy. That gets increasingly difficult as the frequency gets higher. IOW, at audio frequencies it's easy, at 100 MHz it's very difficult.
Twisted pairs, and even coax, will show some impedance changes with things like temperature changes and even their physical orientation. Your hybrid might be balanced by adjusting the compensation part of the network once, but if the pair were moved just a little closer to some other conductor or even bent into a different set of curves, the compensation might have to be adjusted all over again.
I asked my officemate, the RF guy, what kind of bandwidth one could get out of a circulator type hybrid. No dice for baseband signalling, the bandwidth needs to be a much smaller fraction of the base frequency.
We agreed it could probably work if you modulated the 100 mb data onto a carrier, but that carrier would probably need to be up in the microwave range for the bandwidth to be a small enough fraction of the carrier frequency. Get up in the few GHz range and small coax is going to show a lot of attention with distance... not to mention twisted pair becoming impractical to keep uniform enough.
On Fri, 28 Oct 2005 06:04:41 -0700, "Howard Henry Schlunder" wroth:
I was born in 1943, but there are *still* gaps in my experience. That's why I hang around here. I fill in gaps for others and (hopefully) others can fill in gaps for me.
The hybrid being discussed does really have the *possibility* to discriminate signals based on their direction in a line. It's done all the time with RF signals, with hybrid networks that can extract "forward" and "reverse" directional information. And 100 MHz ethernet is just a special example of RF signalling.
Of course, "possible" and "easy" are sometimes worlds apart.
On 28 Oct 2005 17:01:32 -0700, cs firstname.lastname@example.org wroth:
Granted, a circulator might not work. But the OP talked about a transformer based hybrid similar to the ones used at audio frequencies. The audio frequency hybrids *will* handle baseband signals. It's just a matter of adjusting few variables. 8-)
All-in-all a single wire ethernet connection would twice as good as a twisted pair. The single wire could be an optical fiber. Directional couplers for optical fibers are really simple. 8-)
It's just summing, there's no modulation happening as the waves pass there's no reason for them to interact.
at your end od the cable there's two signals present (and they sum.) subtracting the signal you're putting out from the measured signal leaves the unknown received signal.
why is that?
signals don't collide in a cable they travel like waves and just slide past each other (unless the sum is large enough to penetrate the cable insulation etc), "Collision" is a not a Physical link layer event, it's a media access control eevent.
The only reason why collisions were considered bad on coax ethernet is because typical installs had more than two terminals and so with two sending at once the other terminals all got garbled data.
I don't know where you're getting XOR from. cables are ananlog devices they do real sums where two signals are present.