I was having problems with a Gilat Ka band D LNB which I had modified to use with an external reference. I was using a reasonable signal generator at Flo/256 locked to a GPSDO as the reference. However, the constellation plots showed significant phase noise. Ignoring any synthesiser noise addition, the phase noise increases with frequency multiplication by 20 log10 (N) dB where N is multiplication factor. I changed the division ratio in the Gilat synthesiser from 64 to 16 (the mixer doubles again so the actual multiplication is twice those figures). The reference is now 625MHz for a 20GHz LO. The constellations show the reduction in spreading and improvement in SNR and BER on 16E, 21538H, 8PSK 1054ks/s.
In this plot, the magenta trace is before mods (/64) and blue is after the mods (/16). It's hard to read accurately on that plot but the improvement is about 12dB