The test of unplugging WE coils has been repeated (with P4 coil driver installed), having a reference curve taken just before. The result is that the noise decreased a bit at 20-40 Hz (fig 1).
In order to understand if the excess noise is 'common' or 'differential', a measurement in LN1 has been taken. This measurement, rescaled by the ratio between LN2 ans LN1 series resistors, explains well the noise (fig 2), so we conclude that the noise is differential, and an eventual common component cannot be appreciated.
Fig 3 shows an attempt to project back the LN1 measurement in terms of board differential noise. The result is slightly higher than expected, and has to be confirmed.