RS485, multidrop protocol

That "feature" just requires low-level programming and may eat some system resources (a timer) if there is no buffering on the serial port. Other than that, it isn't difficult if you plan for it.

Best regards, Spehro Pefhany

Is that the correct ISO standard? I looked up ISO 3303 and found that it relates to "Rubber or plastic coated fabrics - determination of bursting strength". Nothing at all about HDLC protocols. Perhaps you really meant ISO/IEC 3309:193 "Information Technology - Telecommunications and information exchange between systems, High-level data link control (HDLC) procedures - Frame structures".

Try it on a UART with a FIFO.....

Meindert

The PPP framing is actually a special case of HDLC (ISO 3303) framing. There is an address byte: the first byte after the starting flag which is 0xff in PPP. If single byte addressing is enough for your network, just put the address in there.

It may be worth while to think if the PPP link control protocols (IPCP & co) are really needed in a special network like this. The control protocol is needed if the IP addresses and other network parameters have to get distributed when the link is established.

The main problem in a RS-485 network is the distributed control of the transmit enables to keep bus contention out. If the network is a logical star controlled by a single master station, the timing can be solely run by the master.

The main problem is that hardware receive buffering cannot be used to prevent loss of frame timing gaps. You cannot measure the character timing from the FIFO output.

Please forget Modbus/RTU - it seems to be designed for maximum difficulty to transport. The receive interrupt handler does not know the length of a frame unless is parses all of the commands and sub-options of the frame, and the parsing logic is different for a client (master) and server (slave).

Modbus/TCP seems to be as broken, but in another way.

Been there - done that (gotten bitten by a tunneling Modbus bridge).

"some" resources? It eats tons of resources. It makes it impossible to use the receive FIFO in a UART and thus requires an interrupt for every single byte. That makes it almost impossible to use at high baud rates.

Yes - sorry for the slip. The text was at hand - and I copied wrong (blush!).

It's OK, you are forgiven. I am sure many others would have identified the correct standard easily enough by googling for it. Generally you are below the 3% error rate for keyboard entry so no need to blush too redly.

Modbus master and Modbus point to point slave is trivial to implement, since you can rely on the message contents. However, implementing a Modbus multidrop slave is tricky, since you must adhere to the strict timing.

The simplest environment to implement nearly perfect timing would be to use the QUICC co-processor found on the MC68360/MPC860/MPC2860 etc., but even in this case, the Tx message preamble would have to be set to 2 character times (instead of nominal 1.5 character times), the Tx trailer to 1 character time (nominal 1.5 character times) and the Rx end of frame interrupt at 3 character times (nominal 3.5 character times).

Running Modbus multidrop slave at 115k2 or above with a general purpose hardware, you really need a timing accuracy and interrupt latency of 50 us or less, which the standard Linux kernel is definitely not going to offer.

Paul

Wow, a long thread, Thanks for the inputs, I am following up on the links.

It looks like there is no clear cut winner in getting, say half a dozen machines linked using RS485, a simple master-slave polling scheme, where machines queue up outgoing requests and only speak when spoken to would be simplest. The other candidates seem onerous.

The real trick will be to provide a physical layer that can link to a standard TCP protocol stack.

When I get my linux board I'll start dismantling some of the ethernet drivers, maybe I can "fool" the TCP/IP stack to think it is connected to a perfect (unbusy) network.

I'll revisit slip and ppp then.

Thanks again Ian McCrum MI5AFL. ( actually I just realised AX25 might be a good candidate as well...) University of Ulster

Scott

Ted Wood schrieb:

---8