Yes I terminated the line at the scope's input with a BNC "T" and a 50 Ohms terminator on the other side. What's quite interesting with your TG's performance is the shape of the spectrum at full span. What kind of cable / adapters did you use to connect the TG output to the SA input? And what hardware revision is your DSA815TG? Mine is 00.08 (digital) and 00.05 (RF/FPGA). Please see below the TG spectrum that I recorded. I used two N->SMA adapters and a short (25cm) SMA cable to connect TG output to SA input. For better comparison, I scaled one of the screenshots to the same size as yours. I didn't use normalisation. Strange, isn't it?
Here main board 00.04, RF 00.05, Digital FPGA 00.05, boot 00.01.03 and FW 00.01.15. Device is now exactly 3 years old.
See attached screenshot of the TG from 0-1.5 GHz. I replaced the iffy coax with the same setup as you: two N-SMA adapters and 25 cm good quality SMA cable. This looks much better (the influence of unspecified coax+bnc connectors never ceases to amaze me
)
My scan is more jumpy than yours with a few funny bumps, but for a TG quite useable (plm 1 dB variations, easy to normalize out).
The scans from 1030's are a lot worse it seems, especially below 100 KHz. The 815 is ok-ish above 9 KHz..