CAN bus termination - split with a capacitor and common mode chokes

[Simon]

I usually use a common mode choke ( https://www.we-online.com/components/products/datasheet/744235251.pdf ) and a single termination resistor on a CAN bus.

Lots of designs I see have the termination resistors split into two 60 ohm resistors in serios and a capacitor in the middle to ground. This is apparently for common mode noise rejection, but that is what the CMC is for ?

Do the two solutions complement each other or is it best practice to use both together. I have just found lots of reference to 4.7nF being the value to use and at most 100nF. But most of the designs I have previously seen have used anything from 1µF to 10µF. I realize that this value has no actual effect on the filtering as at all times this capacitor should sit at 2.5V. When recessive both bus lines sit at 2.5V and when dominant the voltages are symmetrical around 2.5V so still 2.5V will be present.

Texas instruments put out an application note about the dangers of common mode chokes but it turns out that they are simply warning that if the CAN bus is shorted to the supply voltage there could be a large back emf spike. Well if you have problems with bus lines shorting to a supply voltage maybe you really need to sort that out rather than blame the CMC for ruining you day.

Maybe the split termination is a more recent development to cope with higher speeds than what the differential inductance of many CMC may allow for?

This seems to be an area where many people are just doing what they always did without question. I don't want to load my designs up with a whole load of parts that I don't actually need.

[Postal2]

...This seems to be an area where many people are just doing what they always did without question....

This is done by the same people who came up with the polarity of the non-magnetic coil in pulse power supply. Do as usual, with two 120 resistors at the ends of the longest line.

[T3sl4co1l]

Consider a transceiver in two states: dominant and recessive.

In the recessive state, the transceiver is open-circuit, and CM voltage is defined as the difference between instantaneous bus CM voltage and local ground.  The receiver is thus susceptible: exceed its CM range and gibberish is received.

When a CMC is used, its voltage drop is small: the impedance at radio frequencies might be a couple kohms, but the transceiver is high impedance as well.  Thus the CMC does almost nothing on receive.  (At high enough frequencies, transceiver capacitance takes over and some filtering value can be had.)

In the dominant state, the bus voltage is either forced high, or the transmitter acts in parallel with another transmitter, and both share bus current to some extent.  Either way, the impedance is low, or at least the voltage difference between bus and transmitter will stay low.  The bus CM voltage is thus forced to local ground (or, well, somewhat halfway between GND and VCC, but referenced to GND is the point).

When there exists a galvanic (say DC or mains-frequency) voltage difference between two nodes, their alternating transmissions will pull the bus CM voltage up and down, thus emitting CM radiation coincident with the transitions into/out of dominant state.  (This might be contained in a shielded structure, cabling, cableways, etc., or it may couple to nearby cables, or it may be exposed to space and radiate per se; I use "emitted" generally here.)

When the bus is long, its capacitance to surroundings, or more generally its CM transmission line impedance (whatever that may be; in general it varies with position along the bus, due to variable proximity to surroundings), is relatively low, and a CMC has some impedance to work against.  Thus transmitter CM emissions can be reduced.

For the special case when a node is near/at the end of the bus, it can be terminated nearby, and that termination resistor can be tapped to obtain a common mode reference to the bus, independent of dominant/recessive transmitter state.

When a termination is tap-grounded (via bypass capacitor) and combined with a CMC, very high CM attenuation is possible (i.e., kohms into 30 ohms), and over a broad frequency range.  Conversely, emissions are reduced, as the node presents a fixed (DC) CM voltage, rather than a dependent one.

Since there are at most two terminators on a bus, this isn't a general solution, but in the special case where nodes cluster only near the ends, or point-to-point communication is used, very high signal quality can be obtained even through a rather mean environment.  It's basically like Ethernet without the transformers, but with enough CM input range that it can handle most anything else (and is also slow enough, and robust to errors, that the remaining stuff -- EFT and ESD most likely -- can be dealt with; something Ethernet doesn't have a luxury of, hence the transformers -- and CMCs).

Tim

[Simon]

So essentially both are a good idea which is what I was dreading for my component count . So midway devices that are not at the end should have a choke anyway. At the moment i am dealing with very short cable lengths although I will be using ribbon which is not the best but with a bus length of under 500mm I figured not an issue. Any cable that would leave the machine would be twisted.

I could say that given the short distance with devices connected by a 150mm cable and the power to these little things being isolated the chokes may be superfluous but for the fact that I will have to attach the ground at one point to a motor driver and um, I don't know if now my ground will be a bit shaky making CMC rather useful.

[T3sl4co1l]

It's worth understanding / appreciating / reiterating that CAN, like RS-485, etc., is a common-ground bus, and not truly differential like we might hope it to be.  (Case in point: CM input range.) A receiver can be fully isolated, without using the tapped-terminator-ground strategy, by finding a (much weaker) ground through the input divider resistors, but this is quickly swamped by pin capacitance; it works at DC and low frequencies, but not over a wide range.

500mm is definitely short enough that everything can be treated as a local node, I guess unless you're using an especially high-speed CAN mode.  Certainly it's fine up to several Mbps.  If those local things can also share ground (in a reasonably low-impedance EMC sense), then a single CMC for the bus entering the local area will suffice, no need for per-transceiver CMCs.

Tim

[Simon]

Well Yes CAN Bus is a funny one with it's 2.5V offset. I guess that at the time it was designed either they just did not think of using a full bridge output or it was not feasible. I am certainly not operating above 1 Mbps.

[Jeroen3]

This is the TI appnote referenced? https://www.ti.com/lit/an/slla271/slla271.pdf

You should remember CAN bus was developed for automotive use, and thus chassis was always near and often referenced.
No need for a fully balanced system. (it would be expensive)

The common mode choke can be used to block some hf noise exiting the node, radiating with the cables.

I don't think you need a choke unless you fail EMC test on the CAN bus wire and can't find the actual emitter.

[Simon]

Yes that one.

[Postal2]

I meant that the coil CMC after the transmitter looks good, when you design the PCB - it should be routed there, but shorted with lines suitable for breaking. But dividing 120 ohms into 2 x 60 with a capacitor is a perversion.

[T3sl4co1l]

I meant that the coil CMC after the transmitter looks good, when you design the PCB - it should be routed there, but shorted with lines suitable for breaking. But dividing 120 ohms into 2 x 60 with a capacitor is a perversion.

Fortunately, EMC doesn't care what you think of it

[Postal2]

From an EMC point of view, it is much more efficient to put a ferrite bead on the pulse power MOSFET, which is usually not done and there is talk about the polarity of the coils.