Bare Ben PCB without LCM.
Unmodified atusd #2, for reference.
Harmonics are absent. (Surprise !)
Signal to "floor": ~20 dB.
Added BK1608HM601 bead to VDD of atusd #1
No significant change.
Added BK1608HM601 bead to CLK of atusd #1
Slight reduction (~2 dB) of noise floor.
Inject clock from a function generator into XTAL1
Square wave Sine wave
Restore host clock, remove CLK bead
Keep 150/1000 pF divider, 10 kΩ, bead on VDD, and 0.5 pF bypass.
External antenna
Cable with 10 dB attenuation, LNA gain reduced by 30 dB
Added 1 μF to EVDD and DEVDD each
No change, as expected.
Added bead to nRST
No change, as expected.
Added 1 μF to AVDD and DVDD each
No change, as expected.
Switched back to resistive clock divider (improved from original)
4.7/1 kΩ followed by 1 nF
Noise spectrum narrows but central "hill" remains unaffected.
Removed 10 kΩ pull-down on XTAL1
(from earlier rework)
Changed divider from 4.7:1 to 5.6:1
Changed DC block of divider from 1 nF to 100 nF
No significant change.
Changed DC block of divider from 100 nF to 10 pF
Slight increase of harmonics.
Reverted DC block to 1 nF, changed divider from 5.6:1 kΩ to 560:100 Ω.
No significant change.
Reverted divider to 5.6:1 kΩ remove bead from VDD
Note: the sidebands at 24505
Testing with external antenna (no anntenuator, gain set back to 46 dB)
Replaced clock divider with bandpass.
Suggested as a last-ditch attempt (before trying a crystal) by Jörg Reisenweber

Commands used in all cases:

atrf-txrx -f 2450 -p 3 -T -0.5  
atrf-txrx -f 2450