We're having some trouble getting a good group delay measurement. It's a frequency converting system and our group delay takes awhile to settle to its final value. The entire trace will shift up and down, sometimes go negative, and will eventually jump a ns at a time to its final value. This process can take a few minutes. We made sure the PNA 10 MHz reference was tied to the system reference and we've played with different IFBW and aperture settings. None of that seemed to help. This seems to only happen when our input signal frequency is greater than the LO frequency. In the other cases, we don't have this issue.