Electronic – How far can a terminating resistor be from the literal end of a CAN bus

Simply put: The old rule of thumb of "you don't need termination if the wire length is less than xx% of the wavelength" is not at all a good rule. There are many cases where it doesn't apply and ISDN is certainly one of them.

The rule is wrong, not ISDN.

Another possibility (which does not rule out what I said above) is that some signaling standards operate on what we call a "current loop". Basically it's the change in current, not the change in voltage, which encodes the data. In these signaling standards the termination resistor is what provides a path for the current. With no current, no data is transmitted.

Update:

Let me elaborate on why I think that old rule is wrong...

1. People misinterpret the rule. They think that "frequency" means the clock or data rate while it really means the highest frequency in the spectrum of the signal. Other people will often say things like "it's not the data/clock rate that matters but the edge rate", which is technically correct but doesn't really help either. The pedantic among us will point out that this is a problem with the person, not the rule-- so sue me.

2. The rule assumes that you know the frequency content of your signal. For digital signals with reasonable edge rates you don't really know this. Further, since good square waves contain an infinite number of harmonics you could interpret this to mean that you always need termination. Of course putting termination on every signal is silly. The point is, for digital signals the rule doesn't give you any useful guidance on if termination is needed or not.

3. The rule assumes that this one signal is the only important signal. It is possible that crosstalk on this signal will couple noise into another signal, and that adding termination (regardless of trace length) will minimize that crosstalk.

There might be other reasons why this rule of thumb doesn't work, but that's all I have right now.

You need to be familiar with Transmission Line Theory to understand the deeper physics in play here. That said, here's the high-level overview:

How important termination is to your system is almost exclusively determined by how long the bus wires are. Here length is determined in terms of wavelengths. If your bus is shorter than one wavelength over 10, the termination is irrelevant (practically) since there is plenty of time for the reflections introduced from an impedance mismatch to die out.

Length defined in wavelengths is a strange unit on first encounter. To convert to standard units you need to know the velocity of the wave and it's frequency. Velocity is a function of the medium it travels through and the environment surrounding the medium. Usually this can be estimated fairly well through the dielectric constant of the material and assuming free-space surrounding that medium.

Frequency is a little more interesting. For digital signals (such as those in CAN), you are concerned with the maximum frequency in the digital signal. That is well approximated by f,max = 1/(2*Tr) where Tr is the rise time (defined 30%-60% of the final voltage level, conservatively).

Why it's 120 is simply a function of the design limited by physical size. It isn't specifically important which value they picked within a broad range (for example, they could have gone with 300 Ohms). However, all devices in the network have to conform to the bus impedance, so once the CAN standard was published there can be no more debate.

Here's a reference to the publication (Thanks @MartinThompson).