4

I know that multiplexing can be used to combine multiple analog or digital signals over a shared medium. Thinking about it I was wondering why we can't use this technique to avoid collisions in broadcast ethernet communications.

Consider a bus network configuration, where several station are connected together sharing the same network cable (bus): if we make any station communicate using its slot in the multiplexed signal, wouldn't be able to avoid collisions inside the bus (making CSMA/CD and CSMA/CA unnecessary)?

What am I missing here?

3
  • 3
    "What am I missing here?" That it would not be ethernet. Bob Metcalfe created ethernet and rejected TDM, opting for CSMA/CD. In any case, shared-medium ethernet is basically extinct. Modern ethernet implementations use switched ethernet that completely avoids the possibility of collisions.
    – Ron Maupin
    Oct 8, 2022 at 23:06
  • 1
    In order to do multiplexing, you need to split the cable's capacity ("bandwidth", though that's technically the wrong term) between the various transmitters. In many situations you only have one transmitter actually transmitting at a time, this means you're effectively dividing the channel capacity by the number of transmitters. So you'd often lose more than you'd gain. Oct 10, 2022 at 3:29
  • 1
    Look at IEEE 802.4 Token Bus, it is pretty close to what you want, but it, too, is extinct as is IEEE 802.5 Token Ring. Neither of which is IEEE 802.3 Ethernet. What you want simply is a completely different protocol..
    – Ron Maupin
    Oct 11, 2022 at 12:24

3 Answers 3

5

Yes, you could, but then it wouldn’t be Ethernet.

Some (now extinct) protocols did exactly as you describe. But they all require:

  1. A method for synchronizing clocks. How do you know when a timeslot starts?
  2. A method for assigning slots. How do you know you should use timeslot 7?

Ethernet won the race in part because it didn’t need either of these things.

3

I would like to add my two cents, and try to explain the problem from slightly different perspective.

In short, what you are missing here is a question of how to assign TDMA slots to the stations. These are the options that I see.

As was said before, neither of these options would be Ethernet.

Option 1: (obvious) assign fixed slots to stations. This has two significant drawbags:

  1. each station gets a fixed slot, irrespective of whether it is planning to send anything or not. As pointed out in other answers, this obviously leads to pretty ineffective utilization of the bus, as slots cannot be allocated to other stations. Besides is not desired even for circuit-switched networks, and breaks the whole idea of packet-switched networks.
  2. adding more systems to the bus requires reconfiguration of each station on the bus.

AFAIK there are some automotive buses that actually work like this, which has more to do with real-time requirements and the fact that you usually do not add stations in a car. Computing schedules for stations, which has to be done for each car model, is nevertheless a big problem. [I am not sure about this statement, please verify it.]

Option 2: have some kind of "bus master" that assigns slots to stations on demand, i.e., when stations wish to send something. There are two options for this master to work

  1. Polling. Master asks each station whether it wants to transmit and allocates the slot if so. Since you will probably assign a dedicated slot for the master to poll, this will indeed solve the collision problem.
  2. Have stations signal the master when they wish to send packets. This can be done using the same CSMA/* mechanism, which does not completely resolve collisions, but reduce collisions to the signalling frames/packets, that should be small.

Ethernet obviously chose not to have a master. I don't think it is immediately obvious who this master could be. There are however systems which do have masters.

  • WiFi Access Point can be a master, WiFi specifies so called point coordination function which polls other stations. Note that this is not the main mode of WiFi operation (see below).
  • signalling as in item 2 is how phone (cirquit-switched) networks, incl ISDN and GSM allocate time slots to phone conversations. You can check ISDN D-Channel, which is a shared CSMA/CD (for me looks more like CA?) bus between all phones/modems on a line and a special device that is a master. For GSM, look for RACH (random access channel) and its role in call setup.

Option 3: have a distributed slot assignment mechanism. Basically this idea is similar to signalling the master, without the master. The time is divided into reservation time and actual time to send data. During reservation time, each station that wants to send a packet tries to reserve a time slot using CSMA/CD mechanism. If succeeded the packet can be sent. Again, this technically resolve all collisions, but it reduce collisions to small signalling packets.

There are several of variants of reservation ALOHA, which is a modification of slotted ALOHA with the added reservation mechanism. WiFi distributed coordination function works like this, and this is the mode that allows WiFi without access point, and the mode that is actually required.


Now, as pointed out by @RonTrunk, using slotted channel access requires each station to synchronize slot start time. This is, AFAIK, done in almost every wireless network, but this is not trivial. Not having slots leads to much simpler design.

Also, as pointed out by all other answers, Ethernet as a CSMA/CD bus is not used today. It turned out to be suboptimal for I guess several reasons. In wired networks it is easier to have point-to-point links and switches, than a shared bus, at least when trying to achieve large throughput.

1

Combine multiple analog or digital signals sounds like you refer to frequency or wavelength-division multiplexing (FDM/WDM) which are just two forms of multiplexing. Those require some central management of frequencies which wasn't envisioned for Ethernet. Ethernet was designed for simplicity throughout.

Multiplexing has many forms. Some of them arbitrate for the medium, some of those can cause collisions, many don't.

There are two general access methods for Ethernet:

  • half-duplex links with CSMA/CD, largely obsolete - this is a kind of dynamic multiplexing where nodes must arbitrate for the medium (coax bus Ethernet or repeated Ethernet)
  • full-duplex, point-to-point links without arbitration (switched Ethernet)

The point with classic Ethernet is that it requires no master on the network for coordinating access. Nodes had to make do by themselves.

Initially, there were quite a few competing technologies, some using a token for medium access (ARCNET, Token Ring, FDDI) which is another method of dynamic multiplexing without collisions. Those required some central entity to initiate and watch over the token. Ethernet was cheaper, so they went under.

Modern Ethernet is fully switched with dedicated, full-duplex point-to-point links between nodes. There is no place where collisions could happen.

Less well-known passive optical variants (xPON) use time-division multiplex (TDM) for the upstream links, just for completeness.

2
  • 1
    So, if I'm getting it right, since some technologies actually use it, you're telling me that multiplexing (e.g. TDM) can be an idea to avoid collisions, is this right?
    – ela
    Oct 8, 2022 at 22:01
  • 2
    Yes, but those methods cannot be used with half-duplex Ethernet. That uses CSMA/CD. You should note that CSMA/CD is all but extinct and modern Ethernet is full duplex and collision free.
    – Zac67
    Oct 8, 2022 at 22:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.