The satellites do beam forming to target the dishes. For this, the dishes communicate their location, calculated from the satellite signals in essentially the same manner as GPS, to the satellites.
The user terminals use both; there’s a switch to enable the user terminal to ignore GPS if it’s inaccurate (for example in conflict areas where it is often spoofed or denied).
They definitely appear to use beamforming internally. It appears that the satellite uses knowledge of receiver location (or at least receiver density) to select beamforming codebooks.
It's less likely they use this in their mobile coverage, but for their own terminals this is absolutely a thing.
They definitely do beamforming, but it's for tracking the stationary service cells as they pass overhead. You can see them represented in the illustrations in that paper.
The SVs (space segment/satellites) in LEO (low-earth orbit) are going to be moving across the sky quite quickly relative to each GT (ground terminal). This is going to be inducing a pretty significant Doppler shift on the signals. Most of the linked paper is focusing on doing ground-based positioning, using the Starlink constellation as a GNSS-type source and talks specifically about how the Doppler shift at the ground is significant enough that not only does the carrier frequency need to track the Doppler shift but also that the baseband signal will be compressed/dilated.
For the SV, though, the same effect is going to happen. The GT is going to be transmitting back to the SV and the SV's going to need to do Doppler compensation in order to successfully decode the OFDM (orthogonal frequency-division multiplexing) signals from the GT. Throughout a pass, each GT is going to have a different Doppler signature based on its position on the ground relative to the SV. The SVs also need to know their positions in orbit with a high degree of accuracy, especially if they've turned on the SV-to-SV laser-based communication (not sure on that). By taking the SV's known trajectory and the per-GT Doppler measurements and making a couple of assumptions (e.g. GTs are stationary), I'm 99% sure you could solve a maximum-likelihood position for each of the GTs. I think you could do it with a single SV but if you have multiple SVs collaborating on it I suspect you could get a quite accurate solution very quickly.
Edit: one other thought I had while writing that up. If they designed the constellation to work this way, I also think it would be possible for the GTs to pre-compensate for the Doppler shift before transmitting, but I don't think I've ever heard of anyone actually doing that. The tradeoff there is GT transmit complexity vs SV receive complexity. I would love if someone pointed out an example of a system that actually does this, but I've only ever worked with systems that just use something like a PLL/Costas loop to track the Doppler shift without needing to estimate it directly. If they did do pre-correction it would definitely make it harder for the SVs to estimate the GT positions. It would make the GTs significantly more complex though and I would seriously doubt that Starlink would go through that effort to make the GTs more expensive and eliminate the ability to identify where they are on the ground.
I don't think pre-compensation would help here. There's multiple SVs in range at any given time so it's just correlating a shifting frequency signal instead of a fixed signal. Provided sufficient difference in the skew between SVs that should be quite trivial.
Anyway thanks. That definitely answers my question. If anything I'd expect GT location data to be more accurate than the terrestrial cell network, at least if the operator bothers to derive it.
Yeah I guess I was making the assumption that a GT was only talking to a single SV at a time. With multiple SVs involved the pre-compensation idea wouldn't help.
Yes, remember Elon Musk was able to cut off Ukrainain military at the exact movement they crossed into Crimea in Ukraine because he knew exactly where the drones were.
This isn't regular Starlink, it is Direct-to-Cell. Direct-to-Cell uses regular LTE/5G. There is no way for the phone to tell the satellite where it is.
Direct-to-Cell probably can't resolve smaller than the cell. I found mixed results if they use the 25 mi cells or if they use the whole dish, hundreds of miles.
The starlink terminal literally has GPS - so Starlink already knows the location - whether it needs to get more accurate coordinates using beamforming is a question.
