1. This forum section is a read-only archive which contains old newsgroup posts. If you wish to post a query, please do so in one of our main forum sections (here). This way you will get a faster, better response from the members on Motherboard Point.

Communications controller?

Discussion in 'Embedded' started by P.Marek, Jun 23, 2005.

  1. P.Marek

    P.Marek Guest

    Hello everyone,

    I'm looking for a cheap system for communication/measurement purposes.
    I'd expect a few I/O ports, 1 - 4 A/D inputs, and some communication

    I've already looked at PICs, AVRs, and lately the ARMs.
    Now I've got the following problems/questions:
    - I'm expecting to transfer up to 32kBytes (256kBit) fullduplex.
    - USB is no good on longer distances - I'm expecting up to 50m.
    How is CAN or I2C in this respect?
    - USART/RS232 won't work, as it has to be some bus. I'd like to use
    4 wire telephone cables (ground, Vcc, 2 communication), but maybe
    is also ok.
    - Ethernet would be possible (and would provide very much bandwidth
    but there a few controllers available - or at least not cheap ones.
    - I wouldn't mind if the ICs had only 14-20 pins - just power,
    and I/O pins.

    PICs are "just slower" AVRs (a bit overstated, but you get my drift),
    but the development tools are available.
    ARMs have much better performance - but are there Open Source compilers
    or assemblers? Using an embedded linux would be simpler, but does have
    much higher cost than the simpler AVRs with only 64kB Flash.

    Any tips?

    Thanks in advance!
    P.Marek, Jun 23, 2005
    1. Advertisements

  2. With CAN the maximum speed you can reach depends on the maximum cable

    the two CAN cables, CAN_H and CAN_L should be a twisted pair of cables in
    your cable.
    gcc is always your friend.
    =?ISO-8859-15?Q?Heinz=2DJ=FCrgen?= Oertel, Jun 23, 2005
    1. Advertisements

  3. Yes, I'd look at serial. For a bus take RS422, that is
    a pair of RS485 drivers, one for each direction
    connected to the internal UART.
    Each direction is a differential pair. This would take
    6 wires when the GND and +5V are included. At least the
    GND is required for reference. PIC and AVR are 8 bit
    machines, while ARM are a different story. What
    performance do you need ?

    Ah, yes, cheap - that means AVR or PIC.

    Rene Tschaggelar, Jun 23, 2005
  4. On 23 Jun, in article
    At what speed, as that is only a data quantity stated there. If that is
    the 'block' size of data how often is it repeated?
    Cannot comment directly on CAN, I have opinions on I2C for this but
    needs answers to above first.
    That is contradictory, unless you are using V24 (ITU.T) specifications
    of RS232 to drive a modem, the RS232 is just a signalling level. Simplest
    RS232 is - TX, RX, signal ground. Running power (DC) down 50m of cable
    will be 'interesting'.
    Work out what you are trying to achieve EXACTLY and what part of the
    system is doing what THEN think about comms method and what electronics
    where. You seem to be very wooly at the moment as to what you want and
    thinking about technology BEFORE knowing what you are trying to achieve.
    Paul Carpenter, Jun 23, 2005
  5. P.Marek

    P.Marek Guest

    I've already looked at PICs, AVRs, and lately the ARMs.
    Sorry, I didn't make myself clear.
    I need to transfer 256kBit per second in both directions
    simultaneously, in short bursts of about 16-128 bytes.
    I read CAN can do 500kBit/s on 100m - that should work.
    It won't work because I need to connect more than two devices. So there
    has to be some bus arbitration and/or collision detection etc.

    The goal is to have some wall-mounted devices which allows voice
    communication to other, identical devices.
    Maybe my best bet is to use the two remaining wires in analog form, and
    to do just access control to these with some other means.

    But then I'd need more than 4 wires, and some additional electronics
    (analog switches, amplifiers) - I hoped that it would be mostly a
    single-chip solution.

    Thank you for your help so far!


    P.Marek, Jun 24, 2005
  6. P.Marek

    P.Marek Guest

    ARMs have much better performance - but are there Open Source compilers
    Can I make standalone binaries that don't need some runtime environment
    like libc etc.? Does gcc include the startup-code for ARM, so that I
    simply program the I/O ports as needed?

    That would be a good solution.

    Thank you very much!


    P.Marek, Jun 24, 2005
  7. P.Marek

    P.Marek Guest

    A 20MHz AVR may work ok - but I fear that it's cannot be "upgraded" to
    some other purposes (like having a display) because of speed

    Yes, I thought as much.


    P.Marek, Jun 24, 2005
  8. P.Marek

    linnix Guest

    Analog phone wires are graded for 20 to 30 KHz typical signal. Jamming
    a 500 KHz digital signals on it is just wishful thinking. It's
    difficult to transfer more than 100 Kb/s without compression. With
    compression, your memory and CPU resource requirements go up rapidly.

    USB is just a form of I2C.
    Perhaps on short messages. Not for constant data at that rate.
    You are probably looking at mid-range ARMs.
    (from another message)
    Thinking about display also and you are looking at high end ARMs.
    linnix, Jun 24, 2005
  9. P.Marek

    David Brown Guest

    You won't be building a single-chip solution anyway - all off-board
    digital communication uses driver chips of some sort (i.e., if you go
    for CAN and use a micro with a CAN controller, you still need a CAN driver).

    If you are talking about building some sort of intercom system, you'd be
    better off using one of the wires for analogue sound, and a wire pair
    for a simple low-speed (say, 9600 baud) RS-485 bus for control and to
    synchronize which node is powering the analogue line. Trying to convert
    the voice information to digital and sending it along a network like
    this is a very inefficient way to do it.
    David Brown, Jun 24, 2005
  10. Since you are talking about full duplex, this would imply a point to
    point connection.
    Up to 50 m CAN could operate at 800 kbit/s and since the CAN frame can
    only carry 8 net bytes and since the total frame size is slightly over
    100 bits (125 us) you might be able to transfer 6-8 frames/ms or 48-64
    kBytes/s, so this link would not be able to carry 2 x 32 kByte/s,
    since realistically some overhead would be required to check e.g.
    assembly of the individual frames into a single application message.

    Why don't you use RS-485, which runs nicely on two wires (and a signal
    ground or some special termination tricks) and assign one station as
    the master that polls the other station (slave).

    The RS-485 common mode range is -7 V to +12 V, so if the current
    supply return connector is also used as a signal ground, the
    permissible voltage drop in the power supply return would be 7 V and
    assuming equal conductor sizes, the supply voltage wire drop could
    also be 7 V, so if the remote device with internal regulator requires
    at least 8 V, a 24 V power supply at the local site would be

    With long wires and large numbers of station, put all stations in
    series and use 48 V or similar voltages to power the devices. In this
    case it would be easier to implement the data using a current loop
    with all the receivers and transmitters in series over the same 20 mA
    current loop :)
    This is your forth posting and just now you get out with a fundamental
    requirement !!!

    How do you expect any meaningful answers ?

    OK, so you have 10-20 devices (or whatever, since you did not specify
    the number) connected on the same bus, each constantly transmitting
    256 kbits/s to each other, thus the net throughput needed would be
    2,5-5 Mbit/s and adding the overhead, the bus throughput would be
    close to 10 Mbit/s. Even the 10BaseT Ethernet would be hard pressed to
    work with the maximum number of stations.
    What is so special about that ? Assign one station as the master and
    the other as slaves (or use one dedicated unit as master with no other
    obligations and all the other as slaves).

    With Ethernet, the controller does most part of the collision
    detection, even if no master/slave relationship exists, but the timing
    would be unpredictable.

    Paul Keinanen, Jun 24, 2005
  11. What kind of rusty barbed wires are you using as telephone lines in
    your part of the world :)

    When you see such specifications, you should first check who made such
    specification and for what purpose. You should look at the impedance
    levels and maximum allowed attenuation.

    Those figures sounds like trying to run high quality audio at a
    nominal 600 ohm level through a telephone grade twisted pair, in which
    the frequency response is limited by the cable distributed capacitance
    in a voltage feed system.

    However, if you drive the cable as a transmission line system with the
    generator and load impedance matching the cable characteristic
    impedance (about 100 ohms), the attenuation is mostly due to
    dielectric losses due to the wire insulation and thus much less
    frequency dependent.

    Ordinary telephone pairs have been used to feed broadband signals for
    quite a while, video since the 1930's (BBC 405 line out-of-brodcast
    camera feeds from the streets of London), for a few decades to carry
    T1/E1 digital telephone over two pairs (24/30 calls 1.5/2 Mbit/s).
    Although out of specs, even the 10BaseT (10 Mbit/s) works in quite a
    few practical situations. All of these use quite simple envelope
    signals and at least in the video case needs a simple frequency

    More complex modulation methods, such as OFDM, can run to several
    megabits/s at several kilometers (ADSL) and tens of megabits below 1
    km (VDSL). These distances are mostly limited by crosstalk between
    different pairs in a thick cable, which limits the transmitter power
    and hence the allowed attenuation and bandwidth.

    Even if a cable may work at a very limited bandwidth in some
    applications (high quality audio), the same restrictions might not
    apply in a different situation with different requirements.

    Paul Keinanen, Jun 24, 2005
  12. P.Marek

    Dave Hansen Guest


    Either you know nothing about USB, or you know nothing about I2C, or
    you would be just as happy with the statement that "Ethernet is just a
    form of HDLC." Or some combination thereof.

    FTR: The only thing USB has in common with I2C is that it involves
    electrical signals changing over time.


    Dave Hansen, Jun 24, 2005
  13. P.Marek

    linnix Guest

    It's not just the wires. The taps (or phone jacks) are creating little
    inductors everywhere.
    Yes, you are talking about carrier frequencies. But RS232,I2C and/or
    USB don't use carrier frequencies. At 1 MHz, the corners of such
    signals would be terriblely distorted.
    Again, these are broadband signals, not baseband.
    Yes, if you place enough equipments (and money) on it. You can run any
    distance you want.
    But frequency response does matter in simple baseband transmittions.
    Why do you think I2C and USB are limited to a couple of meters? Sorry,
    my mistake to lump them together, but they are in the same class.
    linnix, Jun 24, 2005
  14. P.Marek

    CBFalconer Guest

    Assuming you are using C, you can always write:

    int main(void)

    and put whatever system dependant initialization you need in init.
    The purpose of the usual prequel is to set up such things as the
    malloc arena, the file system, glob command lines, implement
    redirection, etc. All you really need for an embedded system is to
    set up the initial stack and program counter. That probably means
    a few assembly instructions starting from wherever a system reset
    puts you. You can customize the rest in an init function.

    If you look at the normal loading mechanism in gcc, you will see
    that it usually loads a standard module before the main program.
    This does the initialization, and calls main. It may contain or
    call a lot of code. Use the -v switches to watch what gcc does.
    You will probably see some references to crt0.o module.
    CBFalconer, Jun 24, 2005
  15. P.Marek

    CBFalconer Guest

    Sounds like what POTS telephone service has been doing for well
    over a century, using a single pair.
    CBFalconer, Jun 24, 2005
  16. P.Marek

    CBFalconer Guest

    POTS pairs have an impudence (spelling deliberate) of about 600
    ohms at voice frequencies, but become a pretty good 100 or so ohms
    by 1 Mhz. The losses (and thus attenuation per unit length) become
    greater at the higher frequencies. We have been able to resolve
    picosecond timings over several hundred feet in the past (30 years
    ago), as shown by the S/N of PWM modulated audio, and repeaters
    extend this almost indefinitely.
    CBFalconer, Jun 24, 2005
  17. What on earth would length of messages have to do with that? If you
    want to squeeze the highest possible throughput out of a CAN bus,
    short messages are exactly the wrong way of doing that --- per-message
    overhead would kill your performance.

    CAN at 500 kBit/s can give you an average payload data transfer rate
    of somewhere around 250 kBit/s using *long* messages, depending on
    which identifier format you use, and whether you also stuff data in
    the CAN ID or only in the "data" part.

    It'll still be a very tight squeeze for 2 x 128 kBit/s of payload data
    rate, of course. But if you don't need to go all those 100 m, you can
    switch to 1Mbit/s CAN to get even more data through.
    Hans-Bernhard Broeker, Jun 24, 2005
  18. P.Marek

    CBFalconer Guest

    Which, in turn, invokes the "everything is a line" mantra. This
    ties Ethernet, HDLC, USB, I2C, RS485, and the telephone together
    quite neatly. :)
    CBFalconer, Jun 24, 2005
  19. P.Marek

    linnix Guest

    I meant sending short data messages, not short payload on the BUS.
    Anyway, sound like the OP is trying to make a digital
    intercom/telephone on the cheap. Can is not exactly cheap for this
    Perhap if you only have two devices on it and ideal conditions. Your
    assumption is totally error free communications.
    And special CAN connectors and cables, not POTS wires and jacks.

    If you don't spent the money on data processing (compression,
    modulation/demodulation), you will be spending a lot more in
    communication hardwares.
    linnix, Jun 24, 2005
  20. A branch in the transmission line becomes significant only when it is
    more than 1/10 of the wavelength. Even with the total 50 m cable
    length specified by the OP, this would correspond to 500 m wavelength
    and 400 kHz (vf=0.66), thus the bus would definitely need termination
    at both ends. However, branches from the main bus would certainly be
    much less and thus a 1-3 m branch that might be typical for an indoor
    systems, the critical frequency gets much higher.
    The dielectric losses are usually proportional to the square root of
    the frequency i.e. 3 dB/octave (while simple RC attenuation would be
    directly proportional to frequency i.e. 6 dB/octave) so indeed, the
    corners of the square wave would be rounded and thus, the eye pattern
    would be reduced.

    To get something resembling a square wave, you would have to include
    at least the 3rd and possibly also the 5th harmonic of the baseband
    signal. Also note that when using NRZ (which the usual UART signal
    is), the highest frequency when sending the ...01010... sequence is
    f/2, thus if the data rate is 500 kbit/s, the highest base band
    frequency is 250 kHz and the transmission medium should pass the 750
    kHz frequency quite accurately and also pass the 1.25 MHz frequency
    with some fidelity.

    The waveform can become badly corrupted if there are long,
    unterminated stubs (branches), which will mess around the group delay.
    Just wondering what your definition of broadband is :). At least over
    here it has been mainly a political question.

    The video signal is definitively baseband, since it is just a level
    shifted version of what comes out of the camera tube (or whatever toys
    they use these days :). The T1/E1 is some kind of three level pulses
    to avoid any DC components, but otherwise quite baseband. On other
    networks, using some kind of Manchester coding does not IMHO prevent
    calling it baseband.
    In the 1980's use of the word "broadband" would indicate that an RF
    carrier is modulated by the digital data stream, thus ADSL, COFM-VDSL
    and cable modems would fit this definition.
    The primary problem is the voltages used and thus the driver power
    required. Also ground potential issues becomes critical in longer

    Ask yourself, what is the difference between a telephone cable (CAT 3)
    and a CAT 5 cable.

    For the CAT5 cable the characteristic impedance is specified as well
    as turns/m and losses at some specific frequencies (thus dictating the
    isolation material). For a random selection of existing (installed)
    telephone cables, no guaranties are available of their high frequency
    performance. In some installations e.g. 10BaseT will work, but in
    other, it does not work.

    Paul Keinanen, Jun 25, 2005
    1. Advertisements

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.