@rf-loop --
initially I rearranged your setup with my SSA3021X(+) to check my signal source (first picture).
I decided to do this after I tried a Rigol DG4102 directly on the DSA815 but since the figures were 20dB worse than your measurement, I considered it may also be the generator. After that I used an LPFRS-01-10M rubidium oscillator, attenuated to 0dBm -- just to find the results on the DSA815 to be identical for all practical means. Since now the LPFRS was heated up and properly attenuated, I did all further measurements with this (second picture - yellow trace is the DG4102 as source, the purple one is the LPFRS). The two spurs just next to the carrier are at 63 and 198Hz above the 10MHz carrier. These spurs are generated by the DSA815 otherwise they wouldn't have been identically present in the spectra of both sources. They may be the frequencies of the control loops of some internal LO PLLs. So regarding this near-field performance, the DSA is about two orders of magnitude worse than the Siglent.
But things almost reverse if you look further off the carrier. Hence I extended the span to 100kHz in the next two screenshots. I kept the narrow-span trace on the screen since I changed the RBW setting to 100Hz in order to finish the measurement in this lifetime. Here's one peculiarity with the Siglent: If changing frequency and span settings, the Rigol analyzer also stretches, shifts or compresses "frozen" traces so at least the relevant section of them can be compared to active traces. The Siglent instrument just continues to display the frozen trace on the screen as it was recorded so there's no use of them for comparison after frequency or span have been altered. Room for improvement I would say. Anyway, in these two Screenshots the peculiar phase noise distribution of the Siglent analyzer once again becomes obvious: The noise is lowest approx. 3kHz off of the carrier while it increases by round about 15dB 40kHz off the carrier to gradually drop then to an exponential relaxation profile. In contrary to this, the Rigol has a phase noise profile that's more or less directly exponential without any dips close to the center frequency but which is in general more narrow than on the Siglent.
This means the Siglent is clearly better suited for measuring low modulation sidebands in the audio range than the Rigol.
The last screenshot finally shows that the noise level is completely independent of the attenuator setting. The purple & cyan traces were recorded at 10dB attenuation while the yellow trace was recorded at 0dB attenuation. One thing to notice here is that the Rigol tolerates a signal of 0dBm at its mixer input (before its starts "screaming") while the Siglent is limited to -10dBm. This means, the Rigol can tolerate stronger signals elsewhere in the spectrum while monitoring a weak signal. This doesn't affect phase noise performance of course, as it's easy to see in the picture.
@D3f1ant --
I'm almost certain that you won't have luck with the PIC circuitry to reprogram your FRAM if you install the new firmware. I guess the new firmware somehow reorganizes the usage of the FRAM, otherwise the WP mod should have worked. The (current) PIC circuitry will only re-write 8 bytes at address 0x200. If now for example there are separate timers for each option (and there's more than enough unused space for this in the FRAM), the result of reprogramming only the aforementioned eight bytes is unpredictable. Either I would revert to the WP configuration (though I didn't test if in this case, FW00.01.17 will enable the 10Hz RBW feature - but at least you can always go back to 00.01.16 to use the other options) or wait with the update until more information is available. Maybe we'll have to analyze a new DSA that comes from the factory with the new firmware and all the trials enabled. Yet, I don't know which hobby user would buy a Rigol DSA these days when he can have the Siglent for round about the same money...
Cheers,
Thomas