Yeah, 1 pair for signal, one for shielding. Cat3 and Cat5/5e/6 aren't the same - but I think you can practically get up to 1gbps over Cat3 - 10gbps is unlikely.
The "shielding" is done by measuring the difference in signal between two wires in a pair -- that's why twisted pairs are used. There isn't a dedicated pair for this purpose.
You need only one pair to run half duplex 10/100bt (contrary to what's stated above), and two pair to run full duplex 10/100bt. In FD, there's one pair each for tx/rx. 10/100bt are very similar in this regard.
The different cable categories mostly relate to the wire quality and the number of twists in each pair.
> You need only one pair to run half duplex 10/100bt (contrary to what's stated above)
How do you wire your one pair for half duplex so it works with commercially available Ethernet equipment? I'll grant you that only one pair will be active at a time, and that 10base2 works with two wires, but you need two pair for 10baseT-HD. I'm very happy to be wrong, please point me in the right direction.
There's also 10Base-T1S, but it's a new specification (2020), and seems limited to automotive and industrial applications, not generally available.
10base2 is totally different. 10base2 is thinwire coax - as opposed to 10base5, thick ethernet.
With 10baseT, the T stands for "twisted pair," as opposed to coaxial cables. Everything we're talking about here is going to be baseT, for twisted pair.
In 10baseT full duplex there are only 4 of the 8 rj45 pins used. In 10baseT half duplex there are only 2 of the 8 rj45 pins used. I don't remember which pins off the top of my head, but you could easily experiment to figure it out. I'm sure it's using one of the two pair used in full duplex.
You don't need to wire anything differently. When you select half duplex mode in your driver it will start communicating over only 2 pins. This sometimes happens naturally with faulty or broken cables, when it looks like the other pins aren't connected properly.
100baseT is more or less the same as 10baseT in terms of pinouts, but at a higher frequency.
The pinout for full and half duplex is the same, the behavior is just different.
If you get traffic on the RX pins when you're sending on the TX pins, in full duplex, that's great. In half duplex, it's a collission and you stop sending your current packet and send a jam signal instead through the minimum packet time.
In 10base2, the RX check is more nuanced, because it's reading from the same bus that TX happens on, collision is only signalled if the RX differs from TX. In 10baseT, I think (but again, would love to be corrected), the expectation is that an ethernet hub will isolate a sending port from receiving its own transmission to the shared bus inside the hub, and of course switches are switches.
For the simple case of two Ethernet devices connected with a crossover cable, there is no possibility of actual collision, but if they run in half-duplex mode, recieving a packet while sending will be processed as a collision anyway.
If you connect only one pair, you can connect RX to RX and TX doesn't go anywhere, TX to TX and TX goes (subject to collissions), but nobody listens, or Peer A TX to Peer B RX and get a one way connection; peer B may detect link up because it gets link pulses, but peer A won't. I'm not sure what Auto MDI-X does in that situation, but it still won't work.
Autonegotiaton usually doesn't do anything smart about failing pins or poor connections either. If you've got two pair connected, you can autonegotiate (at 1Mbps, with the 10baseT link pulses) to the best connection offered by both ends, which will then fail to work if you negotiated to GigE and only have two pairs, or you negotiated to 100M and the cable is really terrible. It's getting somewhat common for OS drivers to detect this negotiation succeeds/connection doesn't behavior and restart the negotiation with fewer options though.
If I have some time today, I'll make a single pair cable and confirm, but if it were as easy as you said, there would probably be a web page telling people how to do it.
I could swear I remember running over a single pair, but it's been a few decades so I might be wrong about this!
If 10bT really does always use 4 pins then I wonder what would happen if tx+/rx+ and tx-/rx- were shorted? I wouldn't be surprised if there's no page illustrating it because why would you ever want to?
If you've got a burried run of two pair wire that you're running ethernet on, but something happens and one pair goes bad, it might be nice to use the working pair, because trenching is a pain. Or maybe you have two pair, but you really want a phone line and ethernet (and ethernet+voip won't work). Etc lots of cases where rewiring is hard, but you don't have enough wires.
Right you are! Twists for shielding, two pairs for duplex - which means that since the 90s on, you practically always want two pairs pr connector to enable duplex (be it 100mbit or faster). Ed: but probably 4 for gigE.
