TDD base stations require synchronization (around 3µs) to avoid interfering with each others, but this does of course not apply if there is only one BS or if there is any other way to sync them (e.g. SyncE + IEEE-1588v2, G.8275...). See https://docdb.cept.org/download/220ac21f-b44b/ECCREP216.PDF

As far as I know 3GPP2 cdma2000/EV-DO also need GPS because the codes are synchronous (although I am not familiar with those technologies).

Apart from that, 3GPP FDD base stations only require a stable frequency reference, not a time reference (LTE/WCDMA base stations are not connected to GPS !)

There are indications that signals from the US GPS constellation and other GNSS constellations can be received on at least part of the moon; coverage gaps could be closed with just a couple of satellites near the moon.

https://spectrum.ieee.org/tech-talk/aerospace/satellites/can...

Does the GPS protocol or another protocol already handle satellites in arbitrary and arbitrarily large orbits?

It's a problem being worked on, but it should be possible to use the existing GPS satellites on Moon. Also, given it is just being used as a time source and not for navigation, it should be possible to make a moon-based time signal with less hardware units?

https://www.nasa.gov/feature/goddard/2019/nasa-eyes-gps-at-t...

There's only like an extra 25 dB of path loss to the moon than Earth since GPS satellites are at orbits of 20,000 km. Of course the GPS antennas are probably oriented facing the Earth, but you'd still see GPS satellies on the far side (with a bit more path loss). Since GPS has a 43 dB processing gain from the modulation scheme, and thermal noise levels on the moon are similar to Earth (I think?), that means you need -150 dBm or so at the receiver. I believe newer GPS satellites have an EIRP of 50 dBm or so, so receiving GPS signals on the moon might work without even using a high gain antenna. Use of a high gain antenna would probably completely compensate for the increased path loss to the moon.

Low lunar orbits are unfortunately too unstable.

As there will be only one cellular network, inference coordination and timing is relatively easy to solve. You just use PTP. No need to GPSS Everywhere.

What exactly requires time sync? I think it's mainly stuff like inter-cell handover, MIMO with multiple cells streaming downlink to a single phone, these kind of things. It should work as long as the base stations are synced well -- no matter what reference. Or they just drop the features that require good sync.

[Edit: maybe TDD mode? but they could settle for FDD...]

They could use their own timing source, though? I recall back in the day being able to pull timing source off a circuit.

Yes, that's the clock holdover circuit, usually built using a OCXO or double OCXO or rubidium clock. It's good for 4 to 24 hours. Example:

[0] https://www.rakon.com/19-rak-product/rak-product-ocxo