This question came up in another group, and I wasn't sure what the answer should be.

Let's say you've set up a system using an RS-485 communication bus, with differential transmission in both directions over a single wire pair. Then you decide that you want to connect a new unit to the existing network, and it's some distance away. It's going to run off an independent power supply, maybe just a wall wart. You run a twisted pair to carry the signals to the new unit. But is that enough, or should there also be a ground wire to tie the grounds of the two systems together? Or maybe the two grounds should each be tied to earth ground, rather than be left floating, and that would link them to the same voltage level. But is that legal and safe?

 

The whole idea of a differential pair is to ignore differences in GROUND between two nodes. The longer the bus the more problematic a ground becomes because of the IR drop in that long ground wire. The differential drivers and receivers allow for a range of common mode ground differences between nodes up to some specified maximum length and capacitance.

 

Generally there is no danger if not referenced to earth GND, RS485 is intended to operate without it, as the signals are differential.
In a noisy environment it may require it, however.
Max.

 

The usual method of dealing with a long cable is to slow down the data rate so the rising and falling edges can take whatever time they need and the signal is solid at the sample point. For a UART this would be the middle of the bit as measured from the leading edge of the START bit in the 10 or 11-bit frame.

 

You need to keep the relative ground voltages within the common-mode rating of the transmitter-receivers.
Connecting all transmitter/receiver circuits to earth (mains safety ground) should help with that.
If necessary you could use RS-485 opto isolators, such as here, which can tolerate very high common-mode voltages.

 

My luck with RS-485 is you ground the shield only on one end.

The car wash controllers use it and they don't like both ends grounded at two points We ground one end
The wire has 2 pairs a shield and bare wire warped around the shield you tie it to the main controller side ground

 

Early in our industrial control career, we installed an RS485 network in a wire drawing factory, and just used the 2 wires that RS485 uses, no common ground. Well, that was a mistake. The big motors used caused significant IR drops in the mains around the factory, bouncing the grounds outside the common mode range of the driver chips. This was fixed by running signal grounds between all the stations. So, I would never run an RS485 network without a signal ground. Use shielded twisted pair for the network cable. Sometimes, a series resistor in the ground connection is inserted to limit any ground currents in the shield. The common ground is really needed when switch mode supplies are used as they often have a lot of capacitivly couple noise.
That is what we did for the next 20 years or so, with much less in the way of network problems. Make the shield ground connection optional as in some cases, only on end is needed to be connected, but having the common ground connection available is a must.
Also, do not run the network cable in the same tray as the power lines to large VFD driven motors! (we had no control over that installation).

These are very handy and quite cheap...
https://www.mornsun-power.com/html/product/RS-485-Transceiver-Module.htmlBuilt in isolation, but I found the T/R switching sense was inverted on the module used, so an extra transistor was needed to fix that, or a software mod.
On some of my boards, it is a build option as to the normal MAX485 driver or the Monsun isolated version is fitted.

 

I agree with dendad, you'll need to connect the signal ground at both ends. This will keep both of the differential signals within the common mode range they're intended to work.

I've used an isolated DC-DC convertor and an optocoupler to disconnect the RS-485 signal ground from the distant device's system ground. This avoids ground current flow if the near device and distant device have significant ground potential differences (very common in industrial situations). The device in dendad's link is similar to what I designed into my circuit.

Note that the signal ground is different than the shield ground. If you use shielded cable it's typically more effective if you connect the shield ground only one end.

Keep in mind, also, that RS-485 is not a current loop. Current loop communications need only two wires, as they are not measuring potential.

 

Thank you for the replies, everyone. That's a very interesting collection of responses which pretty much covers the issue. I think it can be summarized as "When conditions are good, you can get away without a ground. When it's tougher, giving all the units a common ground becomes more important".

 

can you provide some numbers? what is the distance? what is the baud rate? what is the driver chip?

RS485 is used as basis for all kind of fieldbus technologies such as DeviceNet, ProfiBus, DH+ etc.
higher baud rats and bus lengths are complementary. ProfiBus uses more capable drivers so lengths over 1km are supported (at low baud rate). at high baud rate such as 12MBps distances are considerably shorter but still very reasonable.

 

Panic Mode, the issue isn't baud rate versus distance. It's whether the units at various points along the line all need to have a common ground reference. I wanted to know if everything can float at a voltage determined just by the communication lines.

Incidentally, if you're talking about 1km range, you're probably going to be running wires outdoors. Then lightning strikes become a concern, even if they don't hit the wires directly! I don't know what the economics are for lightning protection versus shifting to a fiber optic link.