Will Can Bus still work with lower impedance cable?

[ITman496]

TLDR:  Can I use 73 ohm impedance twisted pair cable instead of 120 ohm as long as I match the termination resistors to the new cable or will it not work well?

Long story below:

I am working on a can bus network (designing it right now) in my aircraft that is around I'd say 8 nodes and the total length end to end with them sprinkled along the chain is around 20 feet?

The twisted pair will run along the rest of the wiring harness and thus near some power cables with some amount of amps flowing through them but not terribly much.  And pretty steady draw, no real inductive kicks constantly flying around.  There will be an aircraft AM radio but not near the wiring, and the engine is pretty far away too.

I have a lot of things with twisted pair, and one of them is a very hardy, very nice cable is this:

https://www.belden.com/hubfs/resources/technical/technical-data/english/76502TS.pdf

Belden 76502TS.  I was given a surplus spool of it, it be wonderful.

However, I have a question and have different answers across my googling.  The characteristic impedance of this belden cable is 73 ohms.  I know that can bus likes 120 ohms.  Does it matter that much?  Some results from my searching seems to suggest that as long as I use 73 ohm termination resistors on each end, to match the cable, I'll be peachy.

This cable is shielded very well, very robust, and I like it more then regular solid core ethernet cable.  The main reason I want to use it is because of its durability and shielding.  It will be on plugs that get moved around and used so I am hesitant to use solid core unshielded ethernet due to fatigue and whatnot.  Also this nice wire is free and I already have it.

I'd really rather not redo the harness again if it doesn't work, so I figure I'd ask if anyone has any experience using out of spec cable?

My nodes will all be using MCP2515 canbus transceivers powered by arduinos, by the way.  My data rate I believe will be around 125-250kbit/s

Worst case scenario I can buy proper cable but... hey, I'd love to just use this fancy nice stuff I have around. 

[Berni]

Well the CAN transceivers are probably designed to drive those 120 Ohm with the cleanest signal but they won't complain driving lower resistances because CAN works just fine if you have less or more than 2 terminators. The datarate is also not all that high for the distance so it should be fine.

You can always stick a scope on it and see how clean the waveforms look.

[ITman496]

So a good test would be to stick a node on each end of the wire on the spool with max length and see how clean it looks?  And if it looks good on the spool, it will probably work fine?

Thank you! I'll try that tomorrow.

[Niklas]

The CAN bus signals are quite robust and probably ok, perhaps with minor tweaking of the termination networks.
One thing to look out for is common mode conducted emission on the power supply wires at the 125/250 kHz and their harmonics.
Had an issue with common mode emissions from CAN bus in the lab last year. A 2 m long harness for CISPR 25 radiated emissions measurement, made up with 4 wires and no shielding. I forgot to check the material specs of the wire insulation and the thin PTFE dropped the impedance to about 65 ohms due to Er and thickness. Emissions vanished with a new harness made from PVC insulated wire where the impedance was within ohms of 120.

[T3sl4co1l]

Probably reduced bitrate-distance, due to the lower voltage into the lower load impedance, and because PVC is pretty lossy stuff, rounding over the edges more than a high-Q cable would.

You can kinda-sorta recover some edge-rounding-over by using a slightly high termination (say 82 or 91 ohms).  I suppose it might extend bitrate-distance by a similar amount (i.e., some 10% or so), which really isn't important anyway as you may not have such fine control (i.e. if your choice is steps of 250k, 500k, etc.), and also if you're hitting the bandwidth ceiling that closely, you have a fundamental design problem that shouldn't be solved by cable tweaks.

But if you aren't doing huge runs -- namely, if total bus length is much less than a bit time, it doesn't really matter very much where the terminator is, or if you have one at each end.  It's very tolerant.

Yes, CMCs are a good idea.  A lot of devices have these, and parts are cheap and common.  Typically something like 50uH or 1kohm+ (at 20 or 100MHz) is chosen.  Since you have shielded cable, you might also opt to take advantage of that, perhaps instead of CMCs (or as large values) -- mind that grounding a shield at both ends is the only meaningful application of a shield, and that you may encounter galvanic isolation or ground-loop concerns in doing so.  (The shield can at least be RF-grounded through capacitors, to keep signal quality acceptable while avoiding those problems.)

Tim

[ITman496]

@Niklas
That is good info, thank you.  I am not terribly worried about the power supply as nothing switch mode is running anywhere near the can bus wiring.  Its all 12V feeding into other systems via switches, think of it like an automotive harness?


@T3sl4co1l
Thank you so much for all the information!

How does one calculate the bitrate-distance?  Also, what is a CMC?  Common mode choke?  As in the little ferrite thing on cables? Would you put the twisted pairs around that or are you saying to solder a physical component on my circuit boards?

I am very curious as I may switch this cable out for unshielded varieties as it is.. huge.  And I realized that trying to have upwards of two or three of them going into my dsub connectors might be unrealistic.  So I might try to find some stranded cat5e.

Also I am worried about ground loops if I use shielded cable, what kind of capacitance grounding of the shield would you recommend? I assume you mean hard grounding on one side and a capacitor on the other.  What values of a capacitor would you use?

[T3sl4co1l]

How does one calculate the bitrate-distance?

Good question.

Also, what is a CMC?  Common mode choke?  As in the little ferrite thing on cables? Would you put the twisted pairs around that or are you saying to solder a physical component on my circuit boards?

Yes, yes, and yes.

Ferrite beads may be sufficient for cleanup of RFI after using the shield; though that may not matter all that much, as the receiver probably doesn't respond much to RF.  (Relevant RF sources would be radio antennas, of course.  So, I suppose, ~120MHz for the most part.)

For UTP, a relatively large CMC is needed, so you might as well place the component.  These for instance:
https://www.digikey.com/product-detail/en/bourns-inc/DR331-513AE/DR331-513AECT-ND/3193327

Note also that you want to center-tap the termination resistor, and bypass the middle to ground, say with 10nF.  Without this, there's no impedance for the CMC to work against, and it does very little to improve receive quality.  (It will still work for transmit quality, because when the transmitter turns on, it's forcing the bus to +/- supply voltages, less some drop; the bus has a low impedance to ground.  But not when it's in the recessive state.  So, even during a packet overall, it's inconsistent...)

Also I am worried about ground loops if I use shielded cable, what kind of capacitance grounding of the shield would you recommend? I assume you mean hard grounding on one side and a capacitor on the other.  What values of a capacitor would you use?

Right, or caps on both ends is fine too, but no point in leaving metal floating.  Preferably use several in parallel, surrounding the signals, so you can keep the shield going.  For example, if you terminate your shields to a metallic D-sub housing, on the connector use a few caps flanking the ground pins on both sides.  The caps distributed around perform far better than a single cap which will add say 5-15nH in series, and therefore some voltage drop in response to RFI coming up the shield.

Which, if it matters, the terminations can be made a modest distance behind the D-sub shell itself, say to adapt to pigtails that are thin enough to enter the housing.  Wrap the junction with e.g. copper foil tape and solder the joints so it's neatly shielded.  Probably still works fine collecting the shields on lone wires really, but this is also good practice for devices that are sensitive to higher frequency interference (say if you're using plain 74HC gates to hack your own bus, they can sense input changes in the low 100s MHz; or say you're doing an analog signal into some op-amps, the input structures can rectify RF, causing errors).

Tim

[ITman496]

This is why I love this forum. 

Would this be what you are talking about?



Also, will this large choke you sent be what I want value wise for UTP or was this part just an example?  And you mentioned it would be good for UTP, would it hurt if I left it in even if I used cable that did not need it?  Is it one of those things that only makes stuff better?


And noted about the shielding.  I'll definitely keep that in mind if I use shielded wiring.  I've always fought with shielding properly so I'm very happy to learn these lessons even if I don't use them on these wires.  I had no idea that you had to shield both ends and that you could use capacitors to avoid ground loops!  Now I'm very glad to know these things.

[T3sl4co1l]

Yup, that's the complete set of EMC protections.

The choke I linked is a fine value for this, and no it won't hurt on STP either.  A smaller one can be used in that case, or maybe just ferrite beads (higher impedances, maybe around 1kohm @ 100MHz, 1206 chip size) instead of common mode filtering, or none at all.

They also show C_H and C_L which can be used to improve normal mode filtering; these may not be recommended if you have a lot of nodes on the bus (they all add up!), but a few-node or point-to-point bus can make use of them.

Obviously also, a multidrop bus only has the terminators at the ends, which means getting some impedance to ground (to help out the CMC, like with the grounded terminator tap) isn't easy to do in the middle of the run.  In that case, if you have a very noisy environment, you may just need to deal with it, or apply filtering (C_H and C_L) and deal with whatever the maximum bitrate goes to.

Tim

[ITman496]

Makes sense to me! I'll try to design my board to be flexible with what I can use. So I can customize each node to fit where it needs to.  Aka, only add termination resistors to the ones on the ends, for example, and leave the middle nodes unpopulated.  What counts as 'a lot of nodes' on the bus? I will have around 8-10.

Also, a question about the path of the bus.  I have the bus starting at the front of the aircraft and then going out to each wing by the time its done talking to all the nodes in the middle of the aircraft.

Is there a nice way to split a can bus into a T or do I have to go out to one wing, talk to the items there, then go backwards back to the middle and then out to the other wing?  I'm trying if possible to avoid active repeaters and to just have a raw bus, but.. if it must be done, i suppose?  If I can do the convoluted path, is there any harm using other sets of twisted pairs in a UTP cable to go the reverse direction? Or should I keep the cables separate?

[T3sl4co1l]

Stubs are acceptable given restrictions on length and bitrate.  Again, electrical length versus bit time.  If you don't need much bandwidth, it may be worth the weight savings alone (not to mention ditching the heavy cable you have, free though it was!).

You can absolutely use two-pair cable for a loop route; wire the pairs together at the dead end.  Carefully keep track of which pairs are going where!

BTW, are there any restrictions on wiring used in GA craft -- I'm assuming this is the situation, you're working on your own plane?  Commercial craft wiring is skookum as heck, but I'm guessing it's relaxed for GA, as many other things also are.  I've never read up on it, just putting the thought out there.

Tim

[ITman496]

Things are extremely restrictive for GA, which is why I'm not doing that.  I've got an ultralight aircraft, which is the most experimental class of experimental aircraft, with... well, no restrictions what-so-ever.  So I'm putting fancy electronics in because its fun.  It's a minimax 1030R if you are curious!

I consulted my friend who has lots of wire and he offered this up as an alternative:

https://www.alliedelec.com/m/d/c9b490c9eef9b2520977afc5146c0284.pdf

Belden 88641

It does not list its impedance which is a bit worrying, but it sure looks nice and much thinner.  I'm trying to figure out how to calculate that now to see if this will work for me.

[ITman496]

I ran it through this calculator:

https://www.allaboutcircuits.com/tools/twisted-pair-impedance-calculator/

And got a shocking..  14.6 ohms impedance.   

Is it because the FEP insulation is so thin that it brings the conductors too close together?

Parameters used: 

.02011in wire diameter
.012in wire separation (due to wall thickness of 0.006 FEP)
2.1 dielectric constant (looked it up and it said that it was the same as PTFE/teflon, which I googled at 2.1.

Am I messing anything up or this this wire despite being twisted pair and shielded completely unsuitable for can bus?

[T3sl4co1l]

Ah yeah, ultralight, that makes more sense!

You shorted em together.   Separation is center to center, it looks like.  I get... 75 ohms or thereabouts?  Quite typical and reasonable.

Note that STP is lower on account of the shield, making it act more like a pair of coax cables sorta merged.  The above calculation assumes ground at infinite distance, I believe.

Tim

[ITman496]

Oh dear.  I feel like a fool.

Thank you!    

Do you know of a calculator I can use for STP?  I'm just trying to figure out what resistor I'll need...  I guess I can just play with it until I stop seeing reflections, though that is an annoying approach.

Also I meant to clarify, so I am going to hard ground the shield on one end, and the other, capacitive ground to avoid ground loops.   Several capacitors surrounding the ground, what values are you imagining?  0.1uf little smd caps?

Thanks again!!

[T3sl4co1l]

No; get STP that's rated, if possible.  Or measure it.

The figure-eight geometry of the shield (or maybe it's more elliptical, depends!) isn't anything I know of a solution for.  Maybe it has been done, but it's a pretty special case.

Do think I've seen or heard of differential coax, that is, two round wires inside a circular shield.  Might be a calculator or formula for that.

You can always calculate it numerically; ideally a proper field solver is used, but a 2D electrostatic solver like ATLC2 can get approximate results (the result is correct, given the input geometry, but what's missing is the dynamics: the skin effect and loss of the wire and dielectric, dispersion (change in velocity factor versus frequency), etc.).

But anyway, that's a lot more work than is necessary here -- it's going to be somewhere around 70 ohms for STP and 100 ohms for UTP.  Just based on the materials and dimensional ratios of commercial pair.  Better than 20% matching really isn't necessary.

Yeah, caps probably over 1nF is a good idea, 10 or 100nF is fine.

Also, if the remote device is itself isolated from ground, no need to isolate the shield, just tie everything in.  So, uh, that thing's not in a metal structure, right?  All wood and composite or what have you?  Yeah, you're making your own grounding circuit then.  Run power in parallel (same harness and routing), can use the shield for ground return most likely so you only need three wires (VCC, CANH, CANL) and the shield.

Which, can be multiconductor cable once again -- if the VCC wire were, say, bypassed to ground at both ends with capacitors, it would act somewhat like another ground conductor, so your braided multiconductor cable looks suspiciously like braided STP once again, and the same impedance assumption applies.

This breaks down when the cable length is 1/2 wave resonant with the signal, where the VCC line will get some resonant voltage in the middle, and currents at the ends.  There's not really any need for VCC to be hard bypassed -- that was just an illustration.  A ferrite bead at each end (on just VCC) might be better, absorbing any energy coupled to it.  Which, is only coupled from the signals in common mode anyway, so we don't expect that to be a large noise source in the first place, and we don't have any qualms about sinking energy from it (we won't dramatically increase differential signal losses or anything).

BTW, note that ferrite beads saturate easily with current, for example a 100 ohm (at 100MHz) 1206 chip bead might be rated for 2A DC, but its impedance falls by 3dB at merely 100mA DC bias.  These curves can be hard to find (Laird is one brand which does provide them, for almost every part in their catalog).  If you need more current, just use an inductor -- which saturates at a known given current.  The lossiness of the ferrite bead can be emulated with an R or R+L in parallel with the inductor, or an R+C across the connection (VCC to GND).

And then, maybe a TVS from GND to VCC, for example a SMAJ15A or P6KE15A for a 12V bus*, will keep the peak voltage to say 20V or so, even under conditions of inrush, surge, ESD, etc., and that covers pretty much everything you'd want for EMC purposes, both power and signal (as discussed earlier).

*Assuming battery powered conditions.  If this is supplied by alternator, there is a chance of load dump, which is when the battery comes disconnected while the alternator is delivering heavy charge current.  The alternator can't reduce its output that quickly, so the excess energy goes into the system.  Typical waveforms are (for 12V bus), 60V peak, 3ms rise time and 100ms fall time (decaying exponential shape), 2 ohms source impedance.  So, 60V from 2 ohms is quite a lot of peak power if you were unlucky enough to absorb it all at once.  There are many mitigation options available from the automotive world.

Tim

[ITman496]

Oh I have another thread on that alternator debacle, don't worry 

Thank you very, very much for all this advice. I will keep it in mind as I design away.  I may have to shuffle my grounds around a little bit as I originally had a star grounding system from the hub of the power distribution, and obviously my signals do not follow that, but we shall see what I can do!