I do not consider notch filters a good idea. You usually need 10 dBm or more to see the
far-out grass on even good, specialized spectrum analyzers. A crystal will not survive that
very long. A crystal is usually specified for 0.5 mW on the resonant frequency, for long-term
stability even less. 10 dBm is 20 times that level.
If you are serious with bleading edge OCXOs, cross correlation is a must. That allows you
to measure oscillators that are 20 dB better or so than the 2 references that are needed.
You get only the noise that is contained in both measurement channels at the same time.
The ADC and the reference noise is averaged away.
I use a Miles Design Timepod for this. It has the drawback that it is limited to 30 MHz.
I think there are faster/better successors for this, from Microchip and someone else
whose name escapes me right now.
I'm working on a 15 GHz two-channel dual-conversion down converter for it. First IF
is 900..928 MHz with SAW filters, 2nd IF is 0 to 28 MHz. OK, the 2nd IF starts at 5 MHz
because of Mini Circuits ring mixer limitations. And then I cannot get the LT5553 for
the high band, which is unobtainium, so I'll have to use the HMC-220 for now.
If I did not have the Timepod, I'd probably build this machine from Holmes,
which consists mainly from some evaluation boards.
<
http://www.aholme.co.uk/PhaseNoise/Main.htm >
Cheers, Gerhard DK4XP