Phantom knock is quite common on modified engines as the knock sensor location and software calibration usually isn't changed from the OEM install to combat it. Often times it's other sources of noise that get into resonance just in the same frequency range as where the knock sensor 'microphone' is most sensitive. And if this background/secondary generated noise overlaps an ignition time window that the ECU is looking for knock then it counts it incorrectly.
Tends to show up as knock in fairly 'odd' locations in the map where you would not expect it like part load around 2500 rpm. The engine itself should not be knock-limited in such areas and often you see that adjusting fuel or ignition in these area's has little to no effect on this 'knock'. Like mentioned above, cam/tappet noise can be a contributing factor, but even a 4-(2-)1 manifold can give differing exhaust pulses that can set off such resonances. Anything from the car touching the block that's not secured with clips or is not isloated with flexible couplers and such is also a potential source.
It can be quite informative to use an old-skool acoustic knock kit (basically a headphone, amplifier and contact-microphone you fit to the block) and listen to the actual noise. With the human ear&brain you can often hear quite a distinct difference in noise/tone between real knock and issues like these. Good method to get a feel for something like this an wether it would be safe to disable knock detection (or increase noise margins or reduce sensitivity in that range if possible) once it's clear that there is actually no knock present.
Bye, Arno.