How do you calculate loss due to a soldered coax cable direct to the board?
I have an RG316 cable with SMA connector on one end. But I'm trying to decide if it is best to terminate both ends with RF connectors and then plug this assembly into the board, or just to solder the wire to the board (seems to be common practise with wireless routers).
A lot depends on how you manage the transition. If you use two thru-holes, and solder the coax inner and outer to them, you'll get a capacitive discontinuity from the pads and then an inductive bump from the coax loop. Those will be moderate effects at 5 GHz. It's better to try to make something more like an edge-launch transition. Here's a crude version, using hardline:
The teflon pushed out of the end of the coax a bit when I tinned the shield, and that made my air loops a little bigger. Just that much extra distance affects matching and is quite visible on 20 GHz TDR. I didn't bother to tweak this one, because this circuit works in the 100 ps/3 GHz sort of zone.
A better transition would be a plated slot at the edge of the board, that the coax shield would fit/solder into, with the coax center conductor soldering directly into a microstrip line with no bends.
Semirigid is nice at 5 GHz.
You can learn a lot, fast, with a TDR scope and an x-acto knife.