Does ISOLATED RS485 require a ground?

[Simon]

My assumption is no.

If you are on a common ground with non isolated devices say a system of stations that are all mains earthed so there may or may not be a common reference then rather than have a dodgy ground that would cause issues you run one explicitly.

But if you have stations using isolated drivers or that are definitely guaranteed to not be earth referenced then I don't see the need. There cannot be a ground that will give big common mode differences and as the signal is differential in polarity then one being more positive than the other has no reference to ground like any H bridge driver.

[mikeselectricstuff]

It depends.
Although in principle you don't, that assumes perfect common-mode rejection, and that the reference for the  line drivers/receivers will find a sensible level.
The only time I'd use RS485 with no ground reference is when using passive-input isolators ( e.g. IL610), which are essentially current driven

[floobydust]

It requires a ground, RS-485 is a three-wire system. Typically in non-isolated systems the ground is implied between the two systems, as the power line ground so it appears as two wire deal.
In substations, the transformer monitors (10-50MVA) are RS-485 isolated because the transformer's ground can kick up a kV or more during lightning or ground faults. Need three wires then. No physical connection to the transformer is allowed, or else you get roasted RS-485 transceivers.

[Simon]

It depends.
Although in principle you don't, that assumes perfect common-mode rejection, and that the reference for the  line drivers/receivers will find a sensible level.
The only time I'd use RS485 with no ground reference is when using passive-input isolators ( e.g. IL610), which are essentially current driven

I'm looking at a battery application and sending RS485 alongside the short power cable but I would rather not have my ground reference screwed by a spike in power draw and the transient in the negative supply although there would be bypass capacitors in the load. So it occurs to me that if I used and isolated transceiver at one end the bouncing ground reference not would affect it. a 4 pin connector is easy, a 5 pin connector becomes a 6 pin connector often.

[T3sl4co1l]

How much power are we talking?  The CMR is 7V or so, it'll be fine in most cases.

Tim

[Simon]

Only 5A although this is a prelude to bigger things to come (over 100A)

[coppercone2]

someone should draw a circuit because the verdict is argument

for normal RS485 I think of this, single point cable ground on shield and individually grounded devices


https://www.powersystemsdesign.com/articles/isolate-to-communicate/22/5838



dust's post seems to imply the 4th case, where the ground is carried through the shield next to the twisted pair to the other device, I think it would make sense to use a dual twisted pair wire here and use 1 pair for ground and 1 pair for signal.

Is your system shielded?

for automotive I would assume you tie 1 transmitter half and shield to chassis, then float everything else.





the question seems to be if you tie shield to chassis/earth or not. the app note says no the other diagram says yes. That might depend on application, in a substation or car? you don't want to do it, for some kind of tame laboratory environment with good ground you might want to tie it to earth.

[Simon]

Case C: Isolated is what I propose. It's not connected to "the land" in any way as it's mobile. So there is no way of accidentally giving anything a ground that may be dodgy and I don't see the point of giving it a certain ground that will be dodgy.

From what I understand the inputs want to see the polarity across A and B, it's not like CAN bus where they sit around 2.5V so if that changes the recognition of the signal is lost.

[coppercone2]

I am guessing they only recommend connecting the secondary grounds if there is an actual shield that surrounds the conductors, if its just a loose wire connecting the grounds I think that is an antenna then, what would be the function of balancing grounds on two floating circuits?

I don't really understand why they put the balance wire inside of the shield though, in the other case. It might be related to cable capacitance since you suddenly have a capacitance there to the shield that is normally not present.

" Since the receiver ground moves with the signals, the receiver is not taxed with rejecting common mode voltage transients.", this has to do with the common isolated ground, but why is it recommended only with the shield?

[T3sl4co1l]

How is it that they give four, whole, possibilities, none of which manage to use the shield correctly?  Even on the isolated options?  Not sure whether I should laugh or cry...

Tim

[coppercone2]

can you elaborate? the article claims they tested it with 100 feet of wire..
they also claim the optimal network is dependent on transceiver type

[T3sl4co1l]

Stuff like this:



Note that A can have a tapped termination too.  That helps when CM filtering is needed, otherwise there's no impedance to filter against.

B is useless unless the shield is tied both ends.  An open shield just lets in all the CM interference induced along the length of the cable.  You might as well not use it, and indeed, you almost never do (Ethernet is fine even in industrial settings without shielding).

There is a negligible improvement in DM noise, because it's already twisted to cancel out interference.  (Note that situations which break this assumption, like closely spaced cables of equal twist rate, or short spans with proximity to noise sources, will benefit more from the shield.)

The ground loop can be avoided at low frequencies by using lots of bypass caps to achieve a similarly low-inductance AC grounding, and it can be avoided at high frequencies by using ferrite beads on the cable (effectively making it a 1:1 transmission line transformer, matching signal and shield currents).

C is doubly isolated, though it may not need to be; the receiver should use a tapped termination, which will extend CMR significantly.

Single isolation is shown in E.  With grounds shared by shield, a tapped terminator isn't needed, but you can if you want.

D is the same thing but with the shield joining the iso grounds.  May not be permissible, are shields required to be galvanically grounded?  Idunno.  Anyway, with both sides isolated, a ground can be added to the shield anywhere along its length, no problem.

F shows the better version of what they were kinda maybe going for.  The one side shares its ground with shield, no signal quality issues there.  The far side sets the ground reference (like a modification of D), while maintaining the isolated receiver to improve CMRR even further, and using a tapped terminator in support of that.

I suspect this was actually pulled from a LVDS datasheet, what with all the transmitters and receivers being fixed, and the grounding being weird?  Possibly RS-422 but these measures should rarely be needed with the wide input range.  RS-485 is a bidirectional bus so might be shown with transmitter and receiver only for simplicity, or to explain one phase of communications, but in reality both devices are transceivers.

Isolation and shielding are more important for low voltage buses (LVDS, HDMI, USB, etc.), where the CMR is only the supply rail, if even that (all of a volt or two).  Industrial buses like RS-422/485, CAN, etc. have extended CMR built in via input divider resistors, so only need shielding or isolation when the ground loop or interference is expected to exceed that range.  (AC coupled buses like Ethernet, need none at all, as the transformers provide the CMRR required.)

A reminder that, in the above, CMR(R) is frequency dependent.  If you isolation barrier has substantial capacitance (a DC-DC isolator might be from 100s to 1000s pF, less for good low-C or high-dV/dt types, more when you add as much "Y" type capacitance as you like), then for the base case (like, the original 'C' one?), CMR is good at DC, dropping off sharply at the RC cutoff frequency determined by receiver input resistance and isolation capacitance.

This is why the tapped terminator makes such a dramatic improvement.  Instead of 10 or 100kohms between pair and local GND, it can be merely 25 ohms (common mode).

Tim

[David Hess]

If a current flows through the shield because it is connected at both ends, then it will magnetically couple into the signal wires and make the situation worse than if it is connected only at one end.  In a floating application it should not matter because there should be no current.  At high frequencies where the cable is electrically long, then the shield needs to be connected at both ends to block RF which creates a dilemma.  Sometimes the shield is capacitively coupled at one end to block low frequency ground currents from the shield.

https://www.analog.com/media/en/technical-documentation/application-notes/41727248an_347.pdf
https://www.analog.com/media/en/training-seminars/tutorials/MT-095.pdf

[T3sl4co1l]

Check the dot on that transformer, it's in the direction that's helping, not hurting.

It's at low frequencies, where magnetic induction is minor and shield resistance dominates, that ground loop is a problem.  Hence mains frequency buzz in poorly set up audio systems.

The crossover is somewhere in the 10s kHz to low MHz, depending on what type of cable you have.  Leaky cable like RG-58 levels off at some point, corresponding to the shield coverage rate more or less.  Well shielded cable has a lower cutoff frequency, and a deeper asymptote.

https://www.analog.com/media/en/technical-documentation/application-notes/41727248an_347.pdf

Oof.  Well, given the age of the scan it looks practically tube-aged, which is when open shields were mostly practical (tube circuits have high impedance, making electric fields the predominant method of interference; a simple conductor around the connections in question, tied to ground at some point, is all that's needed to address that).  It seems they didn't update that advice even in the age of TTL...

https://www.analog.com/media/en/training-seminars/tutorials/MT-095.pdf

This looks better.  Note that the grounding capacitor is an element in the circuit, and needs to match a shield as well as possible.  A typical example would be a BNC connector with 4+ chip caps radiating out from its ground pins.  This still leaves a lot of gap for CM-DM coupling to take place in, but it's at least better than just one.


Another good cautionary example to cite: lots of people for some reason like worsening their USB by putting something lazy like a single ferrite bead between shield and circuit ground, or a single tiny (10nF?) cap, or something else ineffective like that.  Or sometimes nothing at all (leaving it open).  Always ground (AC at least) the shield!

Tim

[mansaxel]

A conductor (the screen) that is not connected can't conduct. Therefore the screen needs to be tied to something sensible in both ends -- except in those rare occasions where there's so much potential difference that things become dangerous.

The principles of "more grounding, more screening, more interconnection" work, and work well in a very wide frequency range, providing the connections are made in a way that makes them low impedance over all concerned frequencies. The problems in audio systems mostly stem from terminating the screen on the PCB in microphone preamps instead of terminating it in the enclosure.  It's called the "Pin 1 problem" (because Pin 1 in XLR plugs is screen) and was the subject of a very widely quoted paper in the JAES back in 1995[0]. 

JAES being what it is, a stalwart of the time when the IEEE and the Springer mobsters and not the IETF decided how standards are controlled, is hard to find online, but on the Pin 1 Problem web site there are a lot of very readable articles on system interconnection.  I especially recommend the Tony Waldron articles.

[0]

   https://www.aes.org/e-lib/browse.cfm?elib=19105

[ajb]

It depends.
Although in principle you don't, that assumes perfect common-mode rejection, and that the reference for the  line drivers/receivers will find a sensible level.
The only time I'd use RS485 with no ground reference is when using passive-input isolators ( e.g. IL610), which are essentially current driven

I'm looking at a battery application and sending RS485 alongside the short power cable but I would rather not have my ground reference screwed by a spike in power draw and the transient in the negative supply although there would be bypass capacitors in the load. So it occurs to me that if I used and isolated transceiver at one end the bouncing ground reference not would affect it. a 4 pin connector is easy, a 5 pin connector becomes a 6 pin connector often.

Depending on the nature and frequency of the ground offsets you are trying to defend against, there are transceivers with extended common mode ranges that may be sufficient to solve the problem without isolation, and just using the power connection as their ground reference.  The standard range is +12/-7V, but a lot of parts do better than that, and it's not hard to find something like +/-25V operational range and +/-60V withstand range.  If you can expect those common mode transients to be relatively rare you may also decide that some electrical protection on the data lines and message integrity checking are sufficient (and both are a very good idea anyway), and just deal with the occasional corrupted message in software.  However this may not be sufficient if you need to deal with something like a big poorly filtered PWM load that's continuously bouncing the load-side ground reference all over the place (and probably causing you other headaches as well).

As to the fundamental question, it's helpful to look at an equivalent circuit for an RS485 receiver. 



The R1/R2/R3 networks are necessary to attenuate the signal from the full common mode range of the bus into the range of the comparator formed by Q1/Q2.  In the case of a receiver that is isolated with respect to the transmitter, the receiver's ground reference will be influenced through this network in combination with any parallel impedances.  Note that this vaguely resembles the tapped terminator Tim sketched into the C and F topologies, where the two bus signal lines are summed through the resistor(s) to a point related to the receiver's ground, but the resistances here will be substantially higher than the termination resistance.  You could calculate the effective values from the input current specs in the datasheet--lower unit load transceivers will necessarily have higher impedances--and from that you could estimate how the whole system will behave, but in general it probably won't be reliable, as the testing on the website coppercone2 linked indicates. 

A similar situation occurs in the conventional bus biasing network, which will by design have an impedance that is much closer to the desired bus impedance although not quite as low as the center-tapped terminator, and will keep the bus common mode voltage ~centered between the receiver's rails.  It also holds the bus in a valid state when it's not being driven which you may or may not care about (fail-safe receivers are common these days).



So one way or another the receiver ground gets referenced to the driver ground, it's just a question of how directly/at what impedance.  Lower impedance reference paths like a tapped terminator or bias network will generally be a better substitute for a real low impedance ground connection than the very high impedance path through the receivers input dividers.

Note that if you have bidirectional communication, you want to consider the receiver referencing at both ends.  Without a ground connection, if you put a bias network or tapped terminator on only one end, it becomes irrelevant when that end is transmitting, because that impedance is swamped by the driver's even lower impedance, so you're back to just having the path through the receiver's input network (at the other end now) to control the ground reference. 

As an aside, networks with more than two stations can't really use these methods to avoid a bus ground reference, since terminating/biasing all of them using conventional values would load the bus down significantly.  I don't think there's a good way to do a reliable multi-station RS485 network without an actual ground reference (whether that's a dedicated conductor in the signal cable is another question though).  The sort of equipment we do uses RS485 through a bunch of portable mains-powered equipment that can be spread over a fairly large area and with different mains supplies, so for this reason the standard in our industry is for the controller to have a non-isolated biased transceiver with isolated transceivers referenced to the shield in all other devices.  Connector shells are kept isolated from the shield so they don't become live in the event of a wiring fault, but end up electrically connected to the device housing when mated, which is sufficient for our data rates/environments. 

[coppercone2]

The shield connection to the isolated supply side depends on the quality of the isolation power supply I think, maybe they assumed it was bad (transformer capacitance?). I see the floating shield (connected only to the PCB, along with the non floating shield (connected to the PCB and BNC panel connector) in the same equipment in different circuit points for different jacks. So the same fully solid state equipment had a shield that is bonded at two points for some connections and a shield bonded at one point for some other connections, with extra care being taken to do so (it came at a overall decent increase in manufacturing cost and time (with the common financial values of a factory manager) for negligible savings in materials due to how they diligently used heat shrink, follow different procedures, cable storage space, etc. So someone made the distinction to use both types of shield schemes in the same equipment, which likely underwent heavy testing. Those isolated probes dave jones reviewed from that trade show with the weird looking power transformer had quite a bit of engineering go into the isolation power transformer (not the signal path) if I recall... this goes heavy into core type, switching frequency of isolation barrier, etc.

Maybe the app note a safe bet assuming something about the quality of the isolation power supplies (where it generally works with whatever wheras the other schemes you proposed might have problem occasionally with certain equipment), making it too difficult for an app note (better to always work then maximize performance and have people diss your writing because of complex hard to explain issues after a off handed decision to go with the picture that had the most 'bests' associated with it during a quick document scan), and for some reason I see people getting angry at having to use split resistors vs a single resistor for the line termination

[David Hess]

https://www.analog.com/media/en/technical-documentation/application-notes/41727248an_347.pdf

Oof.  Well, given the age of the scan it looks practically tube-aged, which is when open shields were mostly practical (tube circuits have high impedance, making electric fields the predominant method of interference; a simple conductor around the connections in question, tied to ground at some point, is all that's needed to address that).  It seems they didn't update that advice even in the age of TTL...

That application note was published in 1982 or later so well into the solid state integrated operational amplifier era.  I included both to be honest, and the universal solution of grounding the shield at both ends is *not* always the best.  It will be application appropriate.

This is hardly a new problem and I remember running into it with RS-232 connections which had the signal ground and shield tied together at both ends, because ground is ground, right?

This looks better.  Note that the grounding capacitor is an element in the circuit, and needs to match a shield as well as possible.  A typical example would be a BNC connector with 4+ chip caps radiating out from its ground pins.  This still leaves a lot of gap for CM-DM coupling to take place in, but it's at least better than just one.

I have done exactly that in a rather extreme applications.  I needed to block low frequency currents from a ground loop in a power RF application from HF to 1.2 GHz.  I cut the shield of a coaxial segment and soldered surface mount capacitors in to replace the missing shield section and keep the transmission line environment.  Good operation at low frequencies required relatively large value capacitors, but with only those capacitors, and I actually used 16, it completely failed at 1.2 GHz.  I think I ended up with 4 x 100pF, 4 x 1000pF, and 4 x 10,000pf which worked perfectly to a hundred watts, and the low frequency ground loop was practically removed.

I still have that ground isolation cable somewhere for special occasions.  It is about 18 inches of RG-400 with an N-connector on each end.

[poorchava]

It works without ground, as long as CM is kind of in civil range. If it's some horrendously hostile environment (automotive, trains, power industry, CAT IV T&M gear) then ground is mandatory, as the common earth potential can bounce several kV and/or conduct some random kiloamp currents every so often.

[Simon]

It works without ground, as long as CM is kind of in civil range.

that in itself is a contradiction. It either works or does not. My take is that if it needs a ground then why does a motor driven by a H-bridge not need a third ground wire that goes - where?

[RoV]

Galvanically isolated RS485 works well with two wires if bias is applied on all transceivers (like previously shown schematics, using the two resistors to VCC and GND from the two sides of the termination resistor). Those resistors automatically set the common mode of the receiver to VCC/2, while the transmitter doesn't need them. Can become a problem if many slaves are on the same bus, because the cumulated bias may become too much (intermediate nodes will have only bias resistors and no termination, obviously). Bias resistors are normally 2.2 k to 4.7 k.

[mikeselectricstuff]

It either works or does not.

You've never had to debug a flaky RS485 system have you...?

[mikeselectricstuff]

It works without ground, as long as CM is kind of in civil range.

that in itself is a contradiction. It either works or does not. My take is that if it needs a ground then why does a motor driven by a H-bridge not need a third ground wire that goes - where?

Because a motor doesn't care if there is a few hundred volts between the windings and the case. RS485 transceivers, not so much.
The only way to do a truly ground-free RS485 system is using passive input isolators like the IL610, which operate by current and not voltage.

[ajb]

Galvanically isolated RS485 works well with two wires if bias is applied on all transceivers (like previously shown schematics, using the two resistors to VCC and GND from the two sides of the termination resistor). Those resistors automatically set the common mode of the receiver to VCC/2, while the transmitter doesn't need them. Can become a problem if many slaves are on the same bus, because the cumulated bias may become too much (intermediate nodes will have only bias resistors and no termination, obviously). Bias resistors are normally 2.2 k to 4.7 k.

The trick here is that a lower impedance makes for a better reference but at the expense of more load on the bus.  So you have to balance those two factors, and what works in one environment may not work in another, depending on the number/type of nodes, data rate, what the electrical environment is like, etc.  So if you've got two nodes, a tapped terminator or bias network at each end is probably fine, because it's a pretty low impedance but it's also the same impedance as the normal termination you want to have at the ends of the bus anyway.  But if you have a bunch of nodes then you should probably just run the isolated ground reference so you don't have to worry about any of that and all of the standard guidelines for RS485 will apply as normal.

It either works or does not.

You've never had to debug a flaky RS485 system have you...?

Heh, an RS485 link between two nodes on a short cable (assuming short since the OP is talking about running this in a battery loom) may well just 'work or not', but yeah, a larger system with more nodes or higher data rate can display all sorts of weird partial/intermittent failures, particularly when a bunch of nodes are spread out over a network that is relatively long relative to the data rate.  Failures on certain bit patterns, failures that shift between nodes on the bus while you're troubleshooting, and those failures may appear in a completely different part of the network from the physical fault that causes them.

[nctnico]

In the end you need to split the problem in several pieces. First of all the receiver needs to have the inputs within the common mode range. This common mode range is referenced to the ground of the receiver.

Secondly you can use a transformer (for example one for telecom or ethernet use) to make an isolated signal. At the receiving end the transformer will need to be biased (by connecting the center tap to half the supply rail for example) so the receiver has signals within the common mode range. Using a transformer also implies that the signaling used on the RS485 bus is AC only so a UART isn't going to work. You'll need some kind of manchester encoding.