I became suspicious of our PLL noise floor when looking at scattering spectra this weekend. It seems at high frequency we are limited by SR560 noise.
The PLL has been locked at Marconi FM Devn of 10 kHz lately, with SR560 preamp gain of 500. The SR560 has ~10 nVrms/√Hz noise floor, so multiplying by 500 gives around ~5 µVrms/√Hz, which is close to the noise we're seeing above 2 kHz.
Here are two test spectra we took today by locking one Marconi sine wave to the other with our PLL. The input Marconi was at 101 MHz and -17 dBm, around the same stats as our current beatnote.
The dark blue line is the noise floor with Marconi FM Devn of 1 kHz and SR560 gain of 200. The light blue is with Marconi FM Devn of 5 kHz and SR560 gain of 500.
From this plot we can see that SR560 noise is hurting us, not Marconi noise.
Edit awade Tue Dec 5 10:34:23 2017: I think maybe the default settings we were using was Marconi slop = 10 kHz/V and SR560 gain of 200 and the 1kHz/V setting we were upping the gain on the SR560 by factor 10 to maintain the same UGF. But I might be wrong.
Edit awade Tue Dec 5 10:39:01 2017: Also, we need to note the LP settings, I think they were 1 MHz in the case, so unlikely they are important. Might need to test for lower frequency BNs as well, but as it isn't Marconi noise then probably not the issue.