Galvanic isolation - why is it needed?

[Lomax]

I'm designing a simple Can-bus to RS485 converter, and sort of instinctively felt I should use galvanically isolated transceivers, but then as I was struggling to come up with a small, cheap, and simple way to power them and the logic while maintaining isolation I started wondering - what's the point of all this anyway? Outside of my device both networks share a common ground, and a common +24V rail, so will I even be able to achieve any kind of meaningful isolation? I can understand the need when you have mixed voltage rails and some of them are high voltage, like in a VFD for an electric motor, but what's the benefit of the expense and complexity in my case? It's not like my little 5V MCU is going to inject dangerous voltages into the data lines, and if high voltage appears on the supply rail it will surely kill the buck converter anyway? A broken device is just as broken whichever part it was that failed... Wouldn't my device be just as resilient vulnerable with just a couple of opto-couplers on the data lines, and perhaps some clamping on the rail?

[Siwastaja]

If you know that the application will share common reference level, there is absolutely no need for isolation and no benefit whatsoever from doing it, just extra cost and complexity.

CAN and RS485 both can handle many volts of common mode voltage, meaning that even if the grounds shifts (e.g. due to a lot of current in a ground wire which carries supply current), that will be OK. If they shift more than a few volts, you probably have other problems than just communication failing.

But one reason to do isolation is product flexibility. Maybe someone needs them isolated. You can always bypass the isolation, but adding it afterwards is much more difficult.

But e.g. in industrial applications it's very usual to reference voltages to protective earth, and in automotive it's common to reference to battery negative (chassis). In these case isolation is not useful.

[PGPG]

If you go with your wire from one building to the other you should expect several kV voltage difference between your GNDs (pulse during lightning strike nearby) and protect your circuits against it.
If you go with your wire from one building end to the second end you should expect about 50V voltage difference between your GNDs (pulse during lightning strike nearby) and protect your circuits against it (you can do it with isolation or without isolation based on cable resistance and transils at inputs).
If you go with your wire between two devices having common GND and through that GND the Surge pulses can't travel you need not to isolate your connection.

[Lomax]

But e.g. in industrial applications it's very usual to reference voltages to protective earth, and in automotive it's common to reference to battery negative (chassis). In these case isolation is not useful.

The application is a mobile home with a 24 V electrical system, and everything is referenced to battery negative. As you say, common mode rejection is already quite good with both CAN and RS485 - they were designed for this, after all. It will certainly save a lot of money and PCB space (both are important factors here) if I can get away with a single non-isolated buck converter to produce the 5 V, and use that to drive the MCU as well as the (non-isolated) transducers. Total max current draw on the 5 V should be less than 300 mA. I've previously built a Modbus RTU with very basic protection (clamping diodes) for a similar environment and have not seen any failures or crashes. With petrol engines there may be some spikes on the DC feed, but the diodes seem to be enough to take care of this. Might still be worth to put opto-couplers on the data lines?

If you go with your wire from one building to the other you should expect several kV voltage difference between your GNDs (pulse during lightning strike nearby) and protect your circuits against it.

Yes, but that could be done with diodes, no? IIUC galvanic isolation would do nothing on its own to suppress a kV pulse since it would be transferred magnetically anyway? A 1 kV pulse on the primary should produce a 1 kV pulse on the secondary with a 1:1 coupled inductor (and most likely killing it and everything else...)

[Siwastaja]

The application is a mobile home with a 24 V electrical system, and everything is referenced to battery negative.

This is exactly the environment for which CAN was designed for, and RS485 is similar. There is no need to isolate. With isolation you are bringing so much extra complexity to the system that you are more likely to do something wrong and worsen the situation.

But do not forget the signal ground. With CAN it's obvious; it's in the culture of CAN usage, everybody knows it's a three-wire interface. Same applies to RS485, but not everybody understands it. Do NOT use vehicle chassis as signal ground. Use twisted pair for data, and closeby conductor for ground.

And with RS485, don't forget biasing.

[Lomax]

This is exactly the environment for which CAN was designed for, and RS485 is similar. There is no need to isolate. With isolation you are bringing so much extra complexity to the system that you are more likely to do something wrong and worsen the situation.

Thank you! I will carry on without it. Thought it would be fine, but your confirmation is very helpful!

But do not forget the signal ground. With CAN it's obvious; it's in the culture of CAN usage, everybody knows it's a three-wire interface. Same applies to RS485, but not everybody understands it. Do NOT use vehicle chassis as signal ground. Use twisted pair for data, and closeby conductor for ground.

Not sure what you mean; I use screened single pair for the RS485, with the pair going to data A/+ and B/- and the screen going to ground. Should I be doing something else with the screen?

And with RS485, don't forget biasing.

I have already incorporated (switchable) termination resistors on the CAN and RS485 buses, but isn't it the bus master's job to provide bias?

[PGPG]

If you go with your wire from one building to the other you should expect several kV voltage difference between your GNDs (pulse during lightning strike nearby) and protect your circuits against it.

Yes, but that could be done with diodes, no? IIUC galvanic isolation would do nothing on its own to suppress a kV pulse since it would be transferred magnetically anyway? A 1 kV pulse on the primary should produce a 1 kV pulse on the secondary with a 1:1 coupled inductor (and most likely killing it and everything else...)

I was speaking about GND difference that is common mode signal and you are speaking about differential mode signal. You have to understand difference. Nearby lighting strike generated current going through earth is a source of common mode surge. It can be partially converted into differential mode at input/output circuits but differential mode pulse will be much weaker.
20 years ago someone (at net) told me that according to his experience with long RS485 lines (industry environment in USA) if devices have 3kV isolation than from time to time they fail, but if they have 4kV isolation fail not happens to them. Have in mind that if you have 4kV isolation at both ends than you tolerate about 8kV GND difference. Also you should take into account that intensity/frequency of storms varies greatly depending on the region of the world.
The other option I imagine is to have for example 1.5kV isolation barier and short it with 1kV varistor. But I can't say I have a big experience in this matter.

[Siwastaja]

Not sure what you mean; I use screened single pair for the RS485, with the pair going to data A/+ and B/- and the screen going to ground. Should I be doing something else with the screen?

You are doing fine, strongly tie the screen to local GND close to the A/B signals (transceivers) at both ends. I just mention this because leaving the ground connection completely out from RS485 bus is surprisingly usual, you won't believe how commonplace it is, and companies like Texas Instruments show wiring like that on their appnotes.

I have already incorporated (switchable) termination resistors on the CAN and RS485 buses, but isn't it the bus master's job to provide bias?

It is not specifically the master device's job; a well-designed master has a (usually configurable) biasing but unfortunately most designers do not bother and push the biasing down to poor field engineers who might have never heard about it. Modbus standard for example requires that the whole system has biasing somewhere, but doesn't say that products must offer it built-in, just that if they do, it must be documented in the manual.

The only reason why it's logical to have at master is because there can be only one master, and biasing is only needed at one place, so it's kinda sorta logical to put this 1+1 together. Technically it can be anywhere, though.

Never assume a product has biasing; if they don't brag about it they very likely do not have it.

If you are designing a RS485 slave device, then you probably shouldn't be designing biasing to it, but it's not a bad idea to remind your customers about biasing in your manual and refer them to the modbus specification (if your product uses modbus, that is).

[PGPG]

The only reason why it's logical to have at master is because there can be only one master, and biasing is only needed at one place, so it's kinda sorta logical to put this 1+1 together. Technically it can be anywhere, though.

It is not so simple.

I have already written at EEVblog: We gave up biasing about 1999 (when we found MAXIM IC with failsafe specified for A,B lines open and shorted).

But following I have not written yet:
With our assumptions:
- number of devices on bus: 2..100 (used ICs that specify up to 128 (or 256) devices on bus),
- bus length up to 1km,
we were not able to design biasing to be incorporated in master (or even distributed between devices to compensate load added by each driver/receiver) to be good whatever devices set user will collect (assuming only our own devices are on the bus). We just wanted to give the user ready to use solution without expecting him to do any calculations (except how long his bus is).
They have problem even with it. We had a complaining that system doesn't work why the bus was 2km length (along all the corridors (and back) in the hospital).

Using fail save ICs solved that problem for us permanently.