First of all: thanks for taking your time to read this and even more for your thoughts.
@MarkT I'm using the TLE 4905L. It has basically a schmidt-trigger with an open collector output stage.
To get a 1/0 signal, I'm pulling it up with 4,7k to +5V. If the output-transistor is open, the logic 1 is only driven by this pull-up, making it vulnerable to noise, that's why I have put that R/C into the circuit. Without it it is even worse. Judging from scope, the signal after the R/C has pretty much round edges (as an effect of the R/C) but still steep enough to not cause timing problems. This signal is then formed perfectly square by an 7400 (will certainly not go into ‘production’ with that dinosaur
>> And this effect could definitely be fast enough that you wouldn't see it on the oscilloscope at the time scales you are looking at.
That’s my biggest fear. My scope only allows to widening by factor of 10. Even with that magnification I don’t see any spikes. But I’m getting that fuzzy feeling that there are some – just too short to see them. What supports that impression is that the ‚build in software scope’ shows several multiple calls/triggers of the same interrupt - but the subsequent reading of the pin-value gathers the same value as by the call before – which is logically impossible because the trigger is defined as CHANGE. But if it was only an extremely short spike, the subsequent reading of the pin-value would probably not catch this which would explain this behavior.