Connect with us

coax Ethernet

Discussion in 'Electronic Design' started by [email protected], Sep 25, 2006.

Scroll to continue with content
  1. Guest

    Hi all
    I'm trying to solve a major problem we have at work. We have a master
    unit on the surface and multiple sensors in the water. For the past 10
    years they have been networked by RS485 operating at 1.5 Mbps. Our
    production people have gotten the sensors to communicate with the
    surface unit at distances well over 100 m with clever impedance
    matching tricks.

    However, we'd like to switch to Ethernet for robustness and for better
    realization of an ideal "single master, multiple sensors" model.

    Using RS485, we often use a "wet Y" or "wet W" to split the RS485 lines
    into 2 or 3 paths that are connected to sensors a short distance from
    the junction (~1 m or so). I want to figure out how to implement this
    same functionality with Ethernet (either coax or twisted pair).

    We have our cables custom-made. They are armored and quite robust.
    I'd like to avoid putting any electronics inside the Y or W junction.

    For twisted pair, I've heard that you should not unravel the pair more
    than an inch or you'll severely compromise the data integrity.

    I'm hoping that perhaps the old coax approach to Ethernet might be a
    reasonable solution since everything was tied together in an analog
    fashion with BNC connectors. However, it a daisy-chain configuration
    rather than a star configuration, which is really what I'm hoping to
    achieve (perhaps with clever impedance matching techniques?).

    Am I barking up the wrong tree? If I search hard enough am I likely to
    find a solution, or am I just headed in the wrong direction? Are there
    any Ethernet tutorials that you've found helpful at a hardware level?
    Ultimately, I'm going to have to put the whole thing together myself,
    from hardware TCP/IP stack to MAC to PHY, so I'm also interested in
    recommendations for hardware solutions (I'm hoping for at least 5 Mbps

    Thanks in advance for the help.
  2. Todd,
    leaving the end-end terminated transmission line for a T-tapped
    structure is a matter of impedance matching/mismatching.
    Whether twisted pair RS485 or ethernet or twisted pair
    ethernet doesn't really matter.

    I remember the TI RS485 transceivers adapted for long and fast
    transmissions. The 65HVD23D achieves [email protected] and the
    65HVD24D achieves [email protected]

  3. tschoepflin wrote ...
    Perhaps we forget that Ethernet was wired with coax
    (huge RG-8 stuff the size of your thumb, and then RG-
    58, etc.) long before UTP ("Cat5", etc.) ever came along.
    HOWEVER, it was only defined as daisy-chain, never
    as any other topology. For many reasons, impedance
    being only one.

    You didn't mention what your data rates are here?
    If they are low enough you could get away with the
    same kind of CSMA/CD (carrier-sense, multiple-
    access/collision-detect) scheme that Ethernet uses,
    but designed to contend with the kind of reflections,
    etc. you would get from a random tree of coax.

    If you could live with lower throughput, and could
    design your own transcievers, you could get away
    with coax joined any old way you wish, because the
    transcievers would make allowances for the lousy
    transmission medium. But I don't think you would
    be successfull trying to use stock Ethernet interface
    nodes with a random coax tree. Unless maybe
    your the time between packets was MUCH longer
    than the packets themselves.
  4. Guest

    The original Ethernet was not a daisy chain, but a broadcast medium.
    The coaxial cable was supposed to be ternminated at either end, and
    every transceiver transmitted to every receiver on that length of coax.

    Obviously, you can use it as a star system

    The two crucial points about Ethernet were

    1) Collision detection

    2) Random back-off after a collision was detected.

    Collision detection just meant that a receiver monitored its own
    transmitters output, and signalled a collision whenever the received
    signal didn't match the transmitted signal, which immediately killed
    the transmission.

    Random backoff meant that the station didn't try to re-transmit
    immediately, but waited for variable amount of time before attempting
    to retransmit.

    The theory behind this is pretty well understood an works time, but you
    don't want to load an Ehternet link at anything approaching its maximum
    capacity - IIRR around 10% is as high as you'd want to go.

    That is about 1Mbit/sec on the original 10Mbit/sec hard ware, but that
    covered distances of over a kilometre. Waht the 100Mbit/sec hard ware
    looks like I've no idea.

    The single coaxial cable can be extended as an un-rooted tree by simple
    dumb repeaters - bearing in mind that the area that can be covered by
    an Ethernet is limited by propagation delay and attenuation - or by
    more complicated bridges that store and forward packets addressed to
    stations outside a particular net, or packets from outside the net
    addressed to stations within a particular net.

    Either mode of extension allows you to realsie the physical connections
    as a start system, though it does make the system a little more

    It was quite a powerful system, and remarkably simple.

    The is a lot of stuff about it on the web
  5. [snip]

    the poster is talking about a master-slave
    communication system. This means there are
    no collisions.

  6. Guest

    Perfectly true, but he needs to understand that the classical Ehternet
    was not a daisy-chained system, which means that he need some
    understanding of how it worked and why the protocol worked the way it

    If you broadcast Ethernet packets ontp the cable from the master
    station, and only let the addressed station respond, you certainly
    don't need collision avoidance, but the size and structure of the
    minimum Ethernet package is determined by the collision dection system,
    and if he wants to use standard Ehtenet hardware an software as the
    basis of his system he is constrained by these rules.
  7. Guest

    The trickiest Ethernet wiring relates to gigahertz and 100baseT
    systems; if you
    can tolerate 10baseT (10 MHz Ethernet) you won't have that kind of
    at all. For 100m distance, you could use either thinnet (10base2) or
    twisted pair (10baseT) wiring without particular problems, BUT I'm not
    sure the
    exact wire specs for commercial buildings (cat-5) are gonna work for
    a ruggedized application like yours.

    If the wet wiring can carry two pairs with 110 ohm impedance and
    suitably low
    crosstalk, 10 MHz Ethernet will work fine. It might even tolerate
    your RS-485
    wiring. And there were baluns used with 10base2 that allowed
    transmission (this is NOT recommended, and it makes full-duplex
    but... it's a thought). Distance limits and specific losses in the
    wiring are unlikely
    to be critical at 100m.
  8. Joerg

    Joerg Guest

    Hello Todd,

    AFAIK neither coax not twisted-pair LAN are suitable for application
    where you want to just "tie together" cable branches at arbitrary
    points. Depending on what you mean by "multiple" maybe you could
    consider dropping down one pair per sensor.

    Anything else would be more of a custom solution. Can be done though if
    the cables are terminated properly at all end points.
  9. Guest

    Since the classic original 1981 Ethernet was half-duplex anyway, and
    probably perfectly adequate for this application, the
    non-recommendation needn't be taken too seriously.
    Though this does depend on the speed you want and the protocol you end
    up using to get it.
  10. Ethernet is not a 'single master' protocol, its peer-to-peer with all
    the collision detection and other overhead that this entails.
    Its not a daisy chain, its a single bus with impedance matching
    terminations on each end.
    The classic coax ethernet was good for a theoretical 10 Mbps, but 5 Mbps
    might be an optimistic upper limit, depending on the number of hosts,
    packet size, protocol overheads, etc.
  11. Sjouke Burry

    Sjouke Burry Guest

    I have worked with coax Ethernet a lot in the recent
    past. I even have a small 3 computer dos/win311 network
    in my home.
    The big problem with the coax is, that it has a maximum
    length of 75 meters(~80 yards). On one such line, you can
    connect several computers,by cutting the line and putting
    in a BNC T connector.
    At our institute we had concentrator units for 16 cables
    in each corridor,which were then connected to ~ 1 inch
    thick high speed network cable with a high impedance
    T connection. One high speed cable serviced several
    So your 100 m branches would not work well with coax
    Ethernet, unless you could bring those concentrators
    very close to a set of drop down lines.
    If I would be intrumenting this, I would use twisted pairs
    in the sensor cables,you need 2 twisted pairs in each
    one,and connect them to rather cheap router units,
    4 to 16 to a unit.
    Then connect those to a master unit.
    You would need a smart micro down at the end of the
    cable, to interface.
    If you use your data collecting computer as master,
    requesting data from each sensor package in succession,
    there will be no problem with data collision,and data
    traffic can get as high as ~800.000 bytes per second,
    the other 200.000 being needed for the downward
    I really got that speed on dedicated 10Mbit Ethernet,
    with two computers exchanging packets in handshake
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day