As expected I’m even more lost than I was. It turns out the pwm pin is not right next to the CE pin. But I tried turning it off anyway. It didn’t work.
The CE pin is an output from an Atmega328p. I would assume it can drive plenty but that’s where my lack of electronic background shines. So I tried all sort of pull-up and down combinations. Driving the pin low as an output and high as an input with the internal weak pull-up turned on. As an input with a stronger 1k external pull-up, as an output with a 1k pull-down (I didn’t like the 3.3mA shunt to ground when driving high but I had to try) and as an output with a strong pull-up (again, I doubt it could drive more than the pin itself but had to try for completeness’s sake). Nothing worked.
What ‘worked’ in the end was having a 1k resistor on the CE pin from the uC followed by a small length of wire and all of that left… floating. I didn’t really try that as a solution but I noticed it was suddenly working for no apparent reason. With just the wire there is a delay but the packet is still sent. With a 10k it isn’t sent. With 1k it’s practically right away. But it’s friggin’ floating!!! So yeah… I’m lost. I put the NRF w/ PA back in the drawer and maybe one day I will again feel like wasting a bit of time on that. Since it happens on only one board and only with the PA models, I will avoid that combination.
A last note, when it’s not working I can fill the FIFO and all three packets will be sent when I touch the wire (I only have to approach it, I don’t even have to touch it). It really is only the CE pin. Everything else is rock solid.
As a last, last note, I can use a PA model powered with only two coin cells in series with a 3.3V linear regulator and a 33uF cap on the NRF’s power pins and it’s working without a single problem.
Anyway, I’ll keep an eye here but I’m mostly done with that problem. On a final board I would not have those wires picking/throwing up all kinds of crap anyway.