As noticed by the commissioning crew, the phase-jumps were not due to the LNFS's PLL signal level (marginally on the low side), also because this configuration was used in the last years (and during the whole O2-run) without any problem.
Nevertheless today we went in the Injection Electronic Room (IER = "piscina") to measure the 10MHz signal sent from DAQ to the LNFS's PLL input as reference clock.
As visible (photo 1) the signal is a TTL 0-5V signal, not an AC-coupled LV-CMOS as thought.
Plugging it on a 50 Ohm load (AC-coupled scope) the amplitude is reduced a little bit (0-4V), not halved (photo 2), so the source impedance should be "near-zero" Ohm.
Measuring it on a 50 Ohm Spectrum Analyzer (via a DC-block capacitor) give a RF fundamental level of +18dBm (photo 3); considering the wasted power into harmonics, we can assume +20dBm (100mW/50Ohm) of power.
We then prepared a male-female BNC box with a series capacitor (0.1uF) as DC-block and a raw PI attenuator (100 Ohm, 56 Ohm, 100 Ohm); the level on the output is that way reduced to +9dBm on the fundamental, and plugging it on the LNFS PLL input we have over than 3dB (actually measured) of margin before an unlock (was previously 2dB).
The monitored PLL signal level is now increased from 9.9 to 11.6 (photo 4)
Extract from LNFS manual: "Ext Ref SMA input fo r the external 10 MHz reference. This input port has an impedance of 50 ohms. The external reference provided should be at10 MHz +/- 0.1 Hz with a level of +10 to +15 dBm."
We can see (photo 5) the spectrum and level of the signal as sent to the LNFS using the new adapter/box.