Network of MCUs with centralized power supply

It doesn't work that way. The ground and signal wires have different impedances, so the induced voltage will be different. In a differential pair both wires carry the same sort of signal, so the induced voltage will be similar and the noise will subtract out.

There is nothing special about the shield. You just said above you intend to use a ground wire in your cable. Why wouldn't that cause a ground loop if you think the shield can?

But yes, if you have no ground connection on the remote end other than the ground in your cable then it should not cause a ground loop.

Don't forget any power supply (current) noise from each of your nodes. Due to the resistance of the 0 V connector, there are going to be some voltage noise across the cable, adding to the RS-232 serial ground. To eliminate any problems due to the node current noise, RS-485 would be preferable.

For instance DeviceNet uses a two pair cable, one pair carrying the power supply an the other carrying the CANbus data (which is essentially RS-485), so I guess you might find it useful to study DeviceNet wiring conventions.

Is it a good idea to try to re-invent the wheel?

I would probably go for ethernet with power-over-ethernet. The folks who designed that have already considered your worries and use isolation everywhere (both the data and power are transferred over a balanced link and isolated at the device).

Look, for example, at the Arduino line and the Arduino Ethernet Shield with PoE option.

The central node can use an off-the-shelf power-over-ethernet switch.

If the node current consumption is less than 4 mA, you could even use a 4-20 mA current loop directly or in HART style. You would only need a single pair for each node.

Classic RS485 is not a two-wire-only network, a third wire is needed that is variously named reference, shield, ground, earth, common, return or 0V. If you look at typical RS485 chips you will see they have a common mode range of a many volts, meaning the signal carrying pair of wires must remain within that range of the common wire.

By having a central power supply you are making life *much* easier than the situation where nodes are separately powered. Where separate power supplies are used the common connection could be paralleled with all sorts of other pathways, some of which may be electrical equipment earthing straps carrying interference and noise. Hence the quite large common mode range demanded of the interface chips.

It is up to you whether or not your network common anywhere connects to the rest of the planet or building around it. For electrical safety in specific locations some jurisdictions may require the network to be fully earth-free.

If the cabling has a shield then it makes sense to either use that shield as the common (you can think of it as acting as magnetic and electric field shield) or at least connect it to common somewhere (then it is just an electric field shield). If the shield is used in parallel with one of the internal wires, like one side of the pair carrying power, then strictly speaking yes you are creating a loop. But the area of that loop is small, it will not be ideal for low-level signals but for a robust scheme like RS485 it should be fine.

I find it can be helpful to sketch the circuit showing all ground connections as wires - without using any ground symbols. Then imagine all those ground wires have resistance or are cut by magnetic flux and then consider the effects on the circuit.

Happy networking!

piglet

Ethernet has some somedrawbacks in my application:

- higher node cost (compared to RS485 transceivers)

- PCB space

- cabling (star topology with a centrale switch)

- higher central node cost (PoE switch)

- more complex network stack (IP and TCP/UDP)

- more complex MCU

Unfortunately, it's not the case...

you could even use

Of course it depends on the application... using off-the-shelf available hardware can be cheaper than developing your own even when it is more complex. Users may appreciate the use of standardized UTP cabling which is already available, although you could use it for RS485 as well. The use of an off-the-shelf PoE switch avoids the certification of power supplies and the switch can be replaced by a dual-power-supply or secondary DC power type when the customer wants more reliability.

Arduino uses an "intelligent" ethernet chip that already contains the TCP/IP stack so it can interface with the microcontroller at the socket level, which may even reduce your software complexity. (no need to develop link level protocol code)

"plus/minus"... relative to *what*? Are you planning on transformer coupling the data lines to the nodes? Otherwise, how will you ensure you meet the common mode limitations of any balanced (or unbalanced) interface?

Where's the ground? You described "plus/minus" and "A/B". How does "ground" figure into this?

And, can you ensure that they never fail in a way that creates a path to "earth"?

No.

Wires are not "ideal conductors". In a balanced interface, the two complementary signals are sourced similarly and sunk similarly (to each other). The power signals (whether you call them plus/minus or power/ground) are sourced -- and sunk -- differently (than the data signals).

Also, note that a bus configuration means every node sees a different IR drop relative to your central power supply. Remember, ALL of the power dissipated in each node has to come down those power lines so "ground" differs from one node to the next.

Again, are you sure there is no path to earth possible? My incoming telephone (land) line is isolated from earth. Until it fails, over time. Then, the noise floor swamps any signal.

Of course, you can claim its "broken" in that case. Fine. How do you expect things to behave when "broken"? If you encounter a hard short to earth at a remote node and end up with a few volts developed across the "ground" connector, what will happen?

Have you considered how your devices will react to nearby lightning strikes? (100m is an awfully big antenna into each device!)

You've not discussed data rates. You might consider transformer coupling the data lines at each node so you can more safely accommodate a larger common mode range. Even a small audio transformer could suffice.

If you want to be clever, you could also deliver power as AC and alter the frequency/phase to support broadcast data (or timing) FROM the central/master node.

There's a difference between "delivering power over an ethernet connection" and 802.3af. The former is (and has been) a hack that can include everything from "using the spare pairs" to "running DC down the cable" (midspan injector). The latter tends to require a fair bit more kit to ensure compliance with standards.

The power conductors are not isolated -- in either midspan or endspan injectors. It's DC; how do you think a transformer would isolate it?

The data is isolated (transformer coupled) -- as in a regular ethernet link.

Note that you can do the same sort of thing with other interfaces besides ethernet.

PoE switches are considerably more expensive than regular switches. And, tend to have significant limitations on TOTAL power sourced. If, for example, you expected the full ~15W at each PD, you'd find that most PoE switches won't handle more than 1/3 to 1/2 of their ports driven at this level. (even fewer at PoE+).

OTOH, if using just a few watts per node...

But, conforming PSE's tend to expect the PD's to be conforming and not ad hoc implementations (i.e., the PSE wants to KNOW what the load classes for each port are to ensure it doesn't over-commit the power available).

If you opt for an ad hoc "power sourced over ethernet cable" approach, then you can do whatever you want (subject to the current carrying capabilities of the conductors you choose and the insulation on them and any isolation that may be involved).

I know that. But it is standard stuff.

The remote end has a DC-DC converter that converts the 48v to the voltage that the device requires (like 5v) and isolates the device from the power at the same time.

With a "passive PoE" (i.e. just a voltage applied to the "unused pairs" in a 10/100 device) this may or may not be the case, depending on how cheap the manufacturer is.

Sure, but I only suggested it because he might have overlooked this entire idea and went on the path of designing a MCU and interface while he can pull an Arduino off the shelf for nearly no money.

But still, they cost a lot less than a one-off development. So it depends on scale if this is a factor.

A typical PoE switch we use, the HP Procurve J9625A, has 382 watts of power for 24 ports, so nearly 16 watts per port. Lower-end switches may offer a bit less.

All this is available off the shelf for like $40, including an embedded TCP/IP stack. Add an Arduino MCU for $25 and you have a complete working system. (both are 1-piece consumer prices at official outlets)

I assumed that the OP used something like +/-15 V op-amp feed at one point and possibly the midpoint coupled to a single real earth at the power supply.

Since each node is floating, the transceiver potential will be in the middle of the +/-15 V feed, so no problems (i.e close to the real earth potential at the power supply).

True.

As long as each node is floating, this should work fine, as long as you have exactly one grounding point in the system, it should work OK.

You could even skip that single grounding point, but this could cause some static electricity build up in extremely dry conditions, so it might make sense to ground the center point of the central power supply.

No problem for me.

This simplifies a lot.

I would be quite happy to have 500 V isolation between the node signal ground (aka central power supply midpoint potential) and any nearby "grounded" metal objects. Even having the nearby frame accidentally connected to the mains live, the 500 V is sufficient for 120, 230, 400 and 480 V mains. Some regulatory authorities might require 1500/2500 V creepage distance between node signal ground and local metallic ground.

Other than that, I see no problem having a single central grounded (or not) power supply feeding each node.

For protection against direct lightning hit, such systems are flimsy anyway.

It usually makes sense to connect cable shields to case (preferably earth) ground at both ends. That keeps the shield, and the wires inside, from jamming RF and ESD noise into a box.

Safety considerations often require that a box be locally earthed.

RS485 differential would be better than single-ended RS232.

Our policy is: ground everything - boxes, boards, shields, whatever - as much as possible. There's nothing wrong with "ground loops" unless they start a fire.

Actually RS422/485 is a bipolar +/-2 mA current loop, in which the current flows through the 100-120 ohm termination resistor.

The only reason for having a ground path is to bias the receiver differential amplifier (bipolar) transistors. If the receiver would consist of two antiparallel optocoupler, there would be no need for any ground reference voltage.

If you use "fail-safe" termination (those quotes are from the original standard) this will force a receiver floating power supply midpoint voltage close to the RS-485 midpoint voltage and hence bias any receiver bipolar transistor base.

If you have a central power supply and a ground connection, it would make sense to ground the central end of each cable, to avoid static buildup in the shield.

Also look at the cable shield as a unipolar antenna,

That doesn't always work and can cause major smoke but we've had this discussion before.

The power supply isn't dual, there's only 0V and +15Vdc. "Plus/minus" was for +15Vdc and 0V/ground rails. "Minus" is ambiguous here.

No.

I hope ground connection will be sufficient. Many RS485 networks work without opto-coupler and/or transformer on data path.

Ground is 0V of central PSU. What I named "minus" at the beginning.

having a singel centrp PSU is one good way to avoid loops.

yes.

kind of... at 100m you wont get much over 9600 baud on RS232. RS485 should be at-least 100 times better.

You've also got the problem of send being on one wire and receive being on the other which make it tricky to allocate the conductors in a twisted-pair cable... pairing +V with send and ground with receive at the PSU end works best IME.

best not to use RS232 the impedance is all wrong. if you must use single ended do it at a lower impedance.

yeah if you can ignore the local earth, and just use the cable earth, do so.