On all the sites I have found about implementing RS485 networks the schematics always show two wires (A/B) for the data signal, but never a specific ground connection between two stations on the network. Yes, each station is shown with a local ground connection, but nowhere does it actually tell you if a common ground connection is required between stations.
Only on one site which was about galvanic isolation in rs485 networks did it mention in passing that "since rs485 networks require a common ground..." which makes me think that yes it does.
However, an rs485 connection can run for a few kilometres. Ground potentials at the two sites could vary by quite a large amount. Would a ground wire linking the two sites in that case really be such a good thing?
Given the differential nature of the rs485 signal, the signal can be recovered without any reference to a ground as such - the signal is the difference between the A and B voltages, not the different between one voltage and ground - so is ground really needed?
Experimentally I have proved it works (at least over short distances) with a battery powered sender with no ground connection transmitting to a receiver over just the A/B lines.
Yes, shielding is good (and really needed), but should the shield be grounded at both ends or just one?
RS485 is a differential signal so no common ground is required.
Wikipedia says: "Connection of a third wire between the source and receiver may be done to limit the common mode voltage that can be impressed on the receiver inputs."
Without a common ground, the circuit may work, but the energy from the imbalance has to go somewhere and may dissipate as electromagnetic radiation.
I've made 485 circuits without a common ground, and they worked anyway. I've grounded just one end and it worked transmitting shiftout data at 60meters without problems. Maybe I was just lucky
It is a good idea for noise immunity to ground one end... This way if there are ever any potential differences there isn't a fault path through a separate ground wire. Shielding is commonly done in high noise environments. The biggest issue I've seen is 485 without the biasing terminations on either end... It works acceptably well under a MHz but longer wiring (> 30 meters) is a cause for concern at higher speeds or longer cable runs.
Docedison:
It is a good idea for noise immunity to ground one end... This way if there are ever any potential differences there isn't a fault path through a separate ground wire. Shielding is commonly done in high noise environments. The biggest issue I've seen is 485 without the biasing terminations on either end... It works acceptably well under a MHz but longer wiring (> 30 meters) is a cause for concern at higher speeds or longer cable runs.
Bob
All the schematics I have seen (from the likes of Maxim, TI, etc) have only had the biasing resistors at the one end. Are you saying they should really be at both?
The relevant part of that white paper is this, found in the last paragraph:
Rather than using two 620 Ohm resistors at one node, you can also increase the value of the resistors and put them at every node. For instance, if there are eight nodes in a system, you can use 4.7 kilo-Ohm resistors at each node to effectively achieve the same result.
..
It's been nearly 20 years since I deployed an RS-485 network and the National Instruments white paper said what I really meant to say but couldn't find the app notes on.. Thank You Dan.. for helping me clear this up. Memory is the 'second' thing to go with 'middle' age..
Actually, I just did a google search on "RS485 biasing", and there are probably 100 appnotes
from all the leading vendors, and that was just the first one I clicked on, LOL. Shoot, B&B
has an appnote with 130-pages.
All in all, 660 ohms seems a little low to me, but I guess they were wanting low-impedance
to help noise rejection for long lines. But you're burning over 3 mA there. I'd think larger Rs,
in the low-K range, would be perfectly appropriate for shorter runs.
No, The splitting up of the terminators to each 'drop' works much better as it provides a distributed low impedance termination and the reason why I mentioned terminating each drop rather than a single termination to begin with. Each installation is a little different and what works for one may not for another.
Docedison:
No, The splitting up of the terminators to each 'drop' works much better as it provides a distributed low impedance termination and the reason why I mentioned terminating each drop rather than a single termination to begin with. Each installation is a little different and what works for one may not for another.
Bob
You mean biasing - termination should only ever be at both ends and never ever in the middle...
So long as you know the characteristic impedance of the cable accurately enough there is never any need for
termination except matched termination at each end of the run. Indeed attempting to distribute the bias network
across all nodes will lead to reflections at every node and degrade high frequency performance and introduce
phase distortion. It is surely better to use stronger biasing to gain higher noise immunity than by compromising
the transmission-line property?
Of course duplex systems like ethernet avoid the issue by not needing biasing (at expense of more wires).
As regards ground, if using shielded twisted pair the shield should be connected to ground at the sending end only,
assuming a unidirectional link. At the other end it is probably wise to connect screen to ground via a resistor (to
dampen common-mode refections). A lot of this is covered in the 485 documents IIRC.
IIRC Distributed Terminations distribute the task of lowering impedances increasing dc current and having the overall effect of increasing noise immunity at the risk of some speed loss due to cable loading capacitance.... As I tried to point out there are many implementations of RS-485 each slightly different, each a part of a specific instantiation and rather than a flat model are tailored to the job at hand. I don't think I have ever worked on or installed any that were not unique. Some half, some full duplex, some high speed short runs many long run data links and some on Telco leased line copper some on in plant wiring, all with different impedances the trick is to make the data speed as fast as the cable will support within acceptable BER rates or as fast as you can consistent with network characteristics. More than a few with cat 5 wire... and
So long as you know the characteristic impedance of the cable accurately enough there is never any need for
termination except matched termination at each end of the run.
Is quite incorrect the terminations are to both match the cable when the "characteristic" impedance is known or a factor but far more important to center the differential signalling to 1/2 Vsupply for improved noise immunity.
As to knowing the characteristic impedance... Twisted pair/cat 5 - 6?, 70 to 105 ohms balanced apporox. But the major issue with a transmission line is that it must be the same at all points or the data rate slowed until it is the same or deal with degradation of service as nodes are added.
Networked RS-485 doesn't use transmission line it uses twisted pair wiring which was I thought the point of the thread not a system where impedance matching is necessary.
The whole error however stems from my misuse of the word termination because it has two meanings and I wasn't specific. I was however referring to a half duplex relatively low speed network as might be used on an Arduino at perhaps a 100k symbols/sec data rate or a 1MB max data rate @ 10 bits/symbol and the utility of distributing the recommended DC biasing among all the drops. A totally generic system not any kind of high speed data link where everything is impedance matched.
Come to think of it I don't believe I've ever seen that parameter on a box of 4 pair twisted wire cable boxes.
This particular appnote shows TI's conception for how to calculate termination and bias Rs.
It also mentions some of their parts are designed not to require external bias Rs. If you look at page
12 of the datasheet for this part, it shows internal bias Rs with values 9K..36K on the A,B inputs.
@ Dan, I thought the first app note right to the point. great information for both the engineer and simple enough for most to follow.. Just a good grasp of ohms law and basics.
I must note again that what I did was install a couple of RS485 networks for a sprinkler controller I designed in the mid 90's and re design some other golf course controller communication links between master and group controllers or schedulers.... fancy clocks with many outputs.
I wish I had that material and those parts when I was doing my work, It sure would have made things easier.
The biggest problems I had were the effects of lightening strikes. The transverse field can easily be several 10's of thousands of volts and all spark gap and gas tube protected both at the drop and at the console as sporadic damage well. There were infrequent lightening strikes and that I was able to limit to one card as I used an opto-isolated coupling between the comm interface and the processor. Spares were cheap.
Great stuff covered here and the linked app notes and.. even the Digikey link..
A good beginning for anyone to simply get the basics of wiring a half duplex RS-485 Comm Link and making it work. There is enough direct and between the lines information to do a full duplex system. IMO
Conclusion
While the calculation of a failsafe-biased network for legacy
transceivers is straightforward, the use of modern RS-485
transceivers such as the SN65HVD308xE family from
Texas Instruments eliminates external failsafe biasing.
These low-cost devices provide integrated failsafe biasing
for open-circuit, short-circuit, and idle-bus conditions as
well as a rating of 1/8 UL, thus increasing the possible number
of transceivers that can be connected to a bus to 256.
so it's likely that 1990's "legacy" chips did not have the internal bias Rs.
Tons of good information already here. I have a little experience in this area. The environment and application has quite an influence on the 'real world' RS485 behavior. We installed a fairly large and long network of CNC machines on a 4 building campus. We ran RS485 underground where we could, between 'towers' where we had to, and of course to about 100 access points. The "noise" off the CNC machines, grinders, and welders was horrendous. We had to use ferrite toroidal cores to knock down a lot of the noise. 'EVERY connector was shielded and grounded. The cable was shielded, and building to building underground conduit was grounded with 8 foot copper rods every 100 feet. Each CNC machine was grounded to building ground. Was it overkill? The place we installed this system had an outage a month before the rebuild. It has run for over 4 years since the rebuild. We did run into one bit of crazy behavior on one of the remote nodes. We added additional cable to the connection to create a tuned stub to quiet it down. That was fun.
