I am still confused about ground station density. As I understand it, the ground stations and satellites both use beam-forming to direct their signals. Early on I heard conjecture that this would limit the number of ground stations that could be deployed in a given area. At best, this would have required splitting up bandwidth between ground stations in dense areas. Could this account for the lower than expected speeds?
I had actually gone so far as to assume that the early semi-randomness of their beta testing was partly explained by this limitation after hearing countless complaints of "my neighbor got theirs, where is mine?". I also assumed that they were going to make hybridized 5G ground stations for deployment in dense areas.
Those assumptions were dashed when I stayed at a VRBO with a starlink connection, and noticed that both the houses on either side also had their own dishes. I chatted with a neighbor, and we both ran a speed test at the same time, yielding ~90/15.
I tried to do a few months ago some back of the envelope math about Starlink.
They are aiming for 12 000 satellites. A given satellite at the altitude Starlink uses should be visible at any average time from about 3% of the surface of the Earth. That would on average be 360 satellites visible from a given spot if they were all spread out uniformly throughout the sky. But the orbits are such that they don't go far north or south. There will be more visible in the areas they can serve and few to none visible outside those areas. I don't know how big the concentration into the service area is, so I'm just going to guess and say 1 000 visible to a given customer at a given time.
Their next generation satellites are specced at having 80 Gbps bandwidth to the ground, which is enough for 800 simultaneous downloads at the full advertised speed.
But most users do not go all out downloading all the time. I've read that most Comcast users only reach about 1/4 of Comcast's monthly data cap, which corresponds to an average use of 1 Mbps, which is 1/100th of the advertised Starlink speed.
Assuming Starlink users also use an average over time of just 1% of their advertised maximum, we get one satellite could support 80 000 users. Taking into account 1 000 satellites visible at any one time from any one average user location, and the area covered by those satellites, I got that Starlink could handle on average 11 users per square mile or 4.4 users per square kilometer.
Note that this doesn't mean that there can't be spots with higher user density. Just that the average density over the service area would need to not exceed that. Islands of density in a sea of very sparsely populated land would be OK.
My calculations assumes that any user could use any visible satellite. I think they are more constrained than that which probably means my allowable density estimates are a little high.
Whilst I agree with most of what you wrote, you need to adjust your thinking on this bit:
> ... most Comcast users only reach about 1/4 of Comcast's monthly data cap, which corresponds to an average use of 1 Mbps ...
Household internet consumption, on average, varies greatly during the day. Most bandwidth usage is from streaming video, and that is mostly done when people are awake, and for a lot of people only in the evenings.
Your 1% calculation is made with the assumption that usage is constant, and is spread evenly across every minute of the day/month. In reality it is mostly all happening in a six hour period (that's an educated guess). Starlink would have to service this peaky traffic, your numbers are 4x out.
They publish a different number of satellites in every single place I have read. Some places they say 19000, some even more.
Guessing that in places where the cost of keeping the constellation filled out is considered, they quote smaller numbers, and for subscribers per satellite, higher numbers.
I pretty much go by the list on wikipedia, which reflects what the FCC has approved. That total is about 12k by 2027, and half that by 2024. I don't see the FCC approving more units until at least the 2024 benchmark, and they are currently only 40% of the way there.
I had actually gone so far as to assume that the early semi-randomness of their beta testing was partly explained by this limitation after hearing countless complaints of "my neighbor got theirs, where is mine?". I also assumed that they were going to make hybridized 5G ground stations for deployment in dense areas.
Those assumptions were dashed when I stayed at a VRBO with a starlink connection, and noticed that both the houses on either side also had their own dishes. I chatted with a neighbor, and we both ran a speed test at the same time, yielding ~90/15.