The point is that the signal to (unwanted) noise ratio is staying the same over an impressive range of Rx signal levels.
This then is a common mode problem which keeps the levels relative, e.g. it could well be occurring in a mixer, but at some point the noise component is being injected, and probably at an IF frequency. Could the ADF be radiating a small level of LO signal that is then creating an intermodulation probem? I'm not sure if Hangflyer has mentioned that it occurs on only one, or more than one discrete frequencies. We know that it occurs on his local base frequency.
Need to "chew the bone" over this.