I need to improve delay measurements on a network analyzer

[EdwardSMaTeresa]

I work for a company that produces RFoF equipment and recently we were approached to prove our capability to produce a specific item, while we believe we have the ability to create the unit, we lack the capability to directly measure it as proof of results. We have another similar item which works on delay measurements over several kilometers ( with roughly +/- 10 meters of error ).

We now need to measure delay changes from changes at the centimeter resolution, however whenever we try to measure changes of a meter to 3 meters we appear to be about half again further than the actual delay ( 1 meter looks like 1.6849 meter to VNA, 2 meter looks like 3.17 meter and 3 meter is 4.7 meters ). We tried to apply a simple linear prediction model to these numbers but it quickly fell off after 5 meters where it wasnt reading even remotely accurate.

We are currently using a 20 GHz 8720 ES network analyzer, we first perform a through cal at ( 10 GHz with 10 GHz span ) and then connect to the DUT, select Delay format and then complete another thru cal, once we complete the cal we take the read number ( 5.5 nS for 1 meter ) and place the reference at that point which comes out at 1.6849 meters rather than 1.

My co-worker seems to remember there is a correction factor that we could apply to improve accuracy based on one or a couple of the following attributes ;
Frequency Center

Frequency Span

Number of Points

Please help, I've been scouring echatspin echatrandom the internet for a clear guide on this the last 3 days and found nothing.