Capacity of Starlink network

Crucial issue for Starlink is the capacity it will be able to provide to its early users. SpaceX has initially requested FCC for initial license of a million user terminals, but that has been recently expanded to five million, as indication of really strong initial demand. 

In the US alone, main satellite broadband competitors have roughy 1.6 million subscribers (Viasat has 600000 subscribers and HughesNet has 1000000 subscribers). It is highly likely that many (especially premium) subscribers will want to migrate to Starlink due to much lower latency and lower data caps. Both competitors will probably try to fight back by lowering prices and increasing the speeds in their data plans, since upcoming Viasat-3 and Jupiter-3 launches will triple the available capacity of these Internet providers. But the real question is: how much capacity can Starlink offer with its initial constellation?

The basic parameters known: 1600 satellites at 53 degree inclination at 550km orbit. 

According to FCC submitted SPACEX V-BAND NON-GEOSTATIONARY SATELLITE SYSTEM ATTACHMENT A TECHNICAL INFORMATION TO SUPPLEMENT SCHEDULE S document from 2017, satellites will be able to use 35 degree inclination or higher between the user terminal and the satellite. At 550km orbit, each satellite will cover a circle with 750km radius or 1.7 million km2. Note that this information is specified for V-Band constellation (VLEO) at much higher frequency band. 

According to the same document, satellite will support 1.5 degree beam spots. That should give ((90-35)/1.5)2 * Pi = 4000 beams. Now, SpaceX will use the low part of the Ku band (10.7-12.7GHz for communication) on initial constellation for subscriber downlink. That is around 2GHz of bandwidth, which maps nicely to advertised maximum of  1Gbps of (downlink) throughput. 

Now, 4000 beams x 2Gbps would mean 8Tbps throughput per Starlink satellite. That would be more advanced that Viasat-3 (1Tbps) or mPOWER satellites (1.4Tbps), so something is not right there. Real question is how many beams can each satellite maintain at the same time? There seem to be just four phased array antennas at the bottom of the satellite, but it is clear that there is space for up to eight Ku/Ka band antennas in the future. Elon Musk claimed to have around terabit of useful capacity per launch, providing estimate of 16Gbps per satellite, or 4Gbps per antenna. That seems plausible, because due to available power and spot beam focusing, 2 bits per Hz should be easily obtainable. 

The real kicker here is the beam spot size. Assuming 1.5 degree beam radius, that leads to surface spot size of just 14km radius, or 650km2. So on that area, Starlink can provide at maximum 4Gbps downlink bandwidth. Using aggressive 100:1 oversubscription that gives 4000 subscribers at advertised 100Mbps downlink, or roughly six subscribers per km2. Not great, but the competition can support even smaller density. And that is why Starlink cannot be viable competitor for subscriber Internet access in densely or even moderately populated areas.

For initial 1600 satellites, only around 3% will be over the continental US territory at any point in time, or roughly 50 satellites (actual coverage is actually higher because satellites spend more time at higher latitude). Distance between satellites will be over 400km, so user terminals will typically have four satellites in sight. 50 satellites gives less that terabit of capacity over whole US continental territory. With 100:1 oversubscription ratio, that is a million 100Mbit subscribers if they are uniformly distributed across that territory. With 99USD/month, this is just 1.2 billion in revenue. 

Note that the available frequency band is not really a limitation here. Due to phased arrays and quite narrow 1.5 degree beams, SpaceX will not run out of frequency capacity. But it has to scale up the number of phased array antennas in the sky. There are only two ways to make this happen: design and launch bigger satellites, or launch more of them. For the initial constellation, it is easier to launch more of them. 

With F9, SpaceX can probably maintain a launch schedule of roughly 1500 new Starlink satellites per year (roughly billion of CAPEX per year, 25 launches).  With five year lifetime that gives 7500 satellites and roughly five million US subscribers. But technology does not stand still. V-band will enable smaller antennas that will enable packing at least 16 of them in the same form factor. So technology will not stand still.

All this should make SpaceX really keen to push rollout of Starlink around the world.  To support five million subscribers in US, they will have to pack more antennas, provide lower bandwidth tiers (such as 25/3 plans) and simply, launch more satellites beyond the initial constellation size.