Wednesday, December 12, 2018

Some Thoughts About Starlink

Elon Musk is well-known for producing highly optimistic schedules.  That's a common problem with engineers, and we all eventually learn a particular engineer's "personal multiplier":  they give you a schedule, you multiply its length by their multiplier, and that's what you write down in the official schedule.  Musk, being the chief engineer of his companies, doesn't have a manager to apply his own personal multiplier, so his schedules are usually pretty optimistic.

But, in addition to ol' Elon not having somebody to reign in his boundless enthusiasm, his schedules have real, sometimes catastrophic, impact on his companies.  He's already endured two near-death experiences in his storied career as an industrialist:  the first Falcon 1 flight test campaign, where the last launch would have bankrupted SpaceX had it failed, and the ramp up to Tesla Model 3 production.

I think Starlink is going to be his third.

Some Background

First, what the hell is a Starlink?

SpaceX has so far made its money off of manufacturing and launching payloads from Earth's surface to orbit.  By dint of making the first stages of their Falcon 9 and Falcon Heavy launchers reusable, they've dramatically dropped the cost of getting stuff to orbit.

While this "launch services" business is quite successful, the actual number of launches is still fairly modest:  SpaceX will wind up doing 20 or 21 launches in 2018.  I've estimated that, as the F9 Block 5 gets rolled out, with target reusability of 10 launches per core, SpaceX may make as much as $37.6M per reusable F9 launch, and up to $76.9M for each reusable FH launch.  On top of that, they charge extra for certain types of payload services for NASA and the military, so they could easily be bringing in $1B a year off of launch operations.

But SpaceX has huge R&D and capital expenses.  They're busy developing the "Big Falcon Rocket" and its associated "Big Falcon Spaceship" (which have just been renamed to "SuperHeavy" and "Starship", which I hate, so we're sticking with BFR and BFS for this post).  This is SpaceX's Mars-class launcher, and, if it works, it will completely revolutionize the launch business by being fully reusable, capable of lifting up to 100 tonnes to low earth orbit, and being able to rendezvous, dock, and transfer fuel from one BFS to another.  Taking these features together, BFR/BFS may be able to deliver cargo to the lunar surface for about $1000 per kg.  That sounds like a lot--until you realize that the next cheapest platform, the FH, will cost something like $12,000/kg.

[Update, 12/22/18:  John Bucknell in the comments points out that my math on the cost/kg for BFR and FH is hinky, and he's right--if we assume a reusable BFR/BFS that has to get to the lunar surface and back with no refueling on the surface.  But, if you were to deploy a Zubrin Moon Direct-style expendable lander from the BFS in LEO, it comes in at only about $250/kg.  See the comments for more details.  This needs more work.]

But BFR/BFS isn't going to be cheap to develop.  Musk has stated publicly that he expects the BFR/BFS system to cost in the neighborhood of $5B to develop.  That's a tough ask on $1B a year in operating profit.

So SpaceX has decided to become a provider of satellite internet services.  Enter Starlink.

Anybody who's used satellite internet today cordially hates it.  That's due in no small part to the delay:  When your satellite is in geostationary earth orbit, 35,800 km about the equator, the speed of light dictates that the time for a packet to get from you to a server and back is close to half a second. Satellites in GEO make for cheap antennas because they don't have to track the satellite, but they provide a horrible user experience.

Starlink is a constellation of thousands (thousands! but we'll get to that) of satellites in LEO, between 340 and 1300 km in altitude.  That makes the antennas to communicate with the satellites more complicated, but the round-trip time is well below 30 ms.  That's actually quite a bit faster than terrestrial internet, where the round-trip time is dominated by electrical cables, routers, and fiber-optic repeaters.  There are plenty of people who will pay big money to shave a few milliseconds off their round-trip time, especially financial companies, where an edge of a few milliseconds can be worth millions of dollars a day.

On top of that, the entire world is a market for Starlink.  A village in the middle of nowhere can plunk down one Starlink antenna and a wifi network and get vastly better service than the cellular network.

There's reason to think that Starlink can generated multiple billions of dollars in revenue in fairly short order.  It would rapidly eclipse launch services as the principal source of revenue for SpaceX.

But we're talking about a lot of satellites here, and those satellites need to be licensed by the FCC and other government organizations to reserve the electromagnetic spectrum to operate.

Here's where we should probably stop for a moment and consider the size of the constellation that SpaceX is proposing.  In its first phase, SpaceX has a license to launch 4409 satellites.  In the second phase, they plan to launch 7518 satellites.

SpaceX wants to build, launch, and operate 11,927 satellites.

But wait!  There's more!  FCC rules require that licensees be able to bring half of their constellation online within six years of licensing, and the whole thing within nine years.  If they don't, the constellation gets frozen at whatever has been deployed, and the rest of the license is revoked.  It's a reasonable rule:  It prevents people from buying up satellite spectrum and warehousing it for future use.

But the six-year deadline for SpaceX's phase 1 (half of 4409 is 2205 satellites) is March 31, 2024.  And the deadline for phase 2 (3759 birds) is in early November, 2024.

And those dates are where Elon's boundless enthusiasm and optimistic personal multiplier almost certainly guarantee him another near-death experience.

Building the Starlink Satellites

We're talking about an unprecedented number of satellites here.  We're also talking about a suite of technologies that aren't very mature, and which have to work at high scale for the Starlink system to be successful.

There are three biggies here:

  1. Satellite-to-satellite communications links, where the birds are constantly moving with respect to one another.  This is key to Starlink's ability to route packets from the ground at one point, to a satellite near the destination, and back to the ground at the destination.
  2. The actual satellites have to be dirt-cheap, but also reliable enough to accomplish the mission.  A small dollop of good news here:  SpaceX is intentionally limiting the life of the birds, so the usual 15-20 year design life used for GEO satellites isn't a requirement.  They'd probably be happy with 3 years.  That makes things a lot easier.  Still, considering the sheer number of satellites, SpaceX must be planning on building them more-or-less continuously, forever.  By the time they get the constellation fully deployed, the earliest birds launch will be aging out, requiring replacement.
  3. Relatively cheap flat panel antennas for the user ground stations, which have to be able to use phased array technology to track the Starlink birds as they're passing overhead.  Unlike, regular GEO satellites, these puppies don't stay fixed in one place in the sky.
SpaceX reportedly has just fired a bunch of people in their Redmond, WA group, which is in charge of the design and manufacturing of the Starlink.  Rumor was that the management of the team was resisting launch sometime next year.  That's not a super-good sign, but it does at least show and awareness that time is short.

Beyond this, I can't really say how they're going to manufacture enough birds, with enough reliability, to make things work.  But this is exactly the kind of problem that Elon and his engineers (at both SpaceX and Tesla) have proven that they're very good at solving.  Beyond that, I'm simply going to assume that the manufacturing isn't a gating item.

Some Assumptions

The rest of this post is going to examine whether there's a prayer of launching almost 6000 satellites by 2024, so this is a good time to stop reading if you're not interested in the numbers.

As I walk through this, I'm going to make some assumptions:
  1. SpaceX will launch its birds on its own launchers, at their cost.
  2. They'll make full use of both the F9 and FH, but only in reusable configurations.  (That last puts some limits on how big the payload can be.)
  3. The current SpaceX fairing, common between the F9 and the FH, is fairly small:  11 x 4.6 m, with a volume of about 145 m³.  I expect SpaceX to come out with a stretched version of that fairing for the FH.  There's a military standard, EELV Category C, which mandates a fairing that's 16 x 4.6 m, which comes out to about 228 m³.  That's what I've assumed for an FH fairing in all of the stuff below.
  4. SpaceX still has to satisfy their paying customers, so I'll assume that paying customers always get priority over Starlink launches.
  5. SpaceX is not the only launch service provider.  ULA, Blue Origin, and possibly Northrop-Grumman will all have offering in the next 6 years.  I'm assuming that whatever their (fairly modest but non-trivial) demands are on the range facilities of both Cape Canaveral (the Eastern Range) and Vandenberg (the Western Range), these companies will get what they need.
  6. I'm also assuming that SpaceX will eventually launch BFRs from a third range, based in Boca Chica, Texas, down by the Texas/Mexico border, just south of South Padre Island.  SpaceX will have full use of that facility, but there are going to be some issues with the rate at which they can do launches without annoying the environmentalists, the maritime industry, the airlines, and the residents and vacationers on South Padre.

How Big Is a Starlink, and How Many Can You Stick In a Fairing?

The FCC applications describe a Starlink weighing 386 kg at disposal time (which is after all the propellant has been burned), and being 4 m long x 1.8 m wide x 1.2 m deep, without its solar panels, but presumably with its antennae deployed.  You have to make some educated guesses about what this means at launch time.  I'm going to assume four very important things here:
  1. The "wet mass" (with propellant) of the Starlink bird will be 500 kg.
  2. The stowed configuration of the Starlink has the antennae folded across the body of the satellite.  That should make the stowed configuration 1.33 m x 1.8 x 1.2 m.
  3. The birds will actually be trapezoidal prisms, with the 1.8 m with being the base of the trapezoid.  This is pretty standard for small satellites, which cluster in rows around a central "dispenser" cylinder, which is responsible for deploying them into space from the launcher.
  4. I'm going to assume that the mass of the dispenser comes to 75 kg per satellite.
With those assumptions, I figured out how many birds each of the three fairings (current F9, Category C EELV for FH, and BFR) could hold.  That puts an upper limit on how many can be launched.  This was done using the very scientific method of making little satellite models in Powerpoint and seeing how many I could stack into the various fairings.  Here's an example, so you can share my pain:

When I totted up all the various options:, I got the following table:

Starlnks Per FairingStarlink LengthF9 BirdsFH BirdsBFR Birds
Assuming 2.0 m long birds22642232
Assuming 1.3 m long birds1.333962344
Assuming 1.0 m long birds15177500

Note that we're going to be assuming the 1.33 m length below.  If the launch configuration comes out to 2 m, things are even harder than the very hard schedule we'll see below.

The Orbits

The Starlink constellation is organized into a set of "orbital planes".  Each plane contains a certain number of satellites, spaced an equal angular distance apart.  Each plane is inclined to the equator by a certain angle.  Finally, each plane has a specific altitude above the Earth.

Here's an example:  There's a set of orbital planes that are 550 km altitude x 53° inclination.  Each plane has 66 satellites in it, which means that they're spaced 5.45° apart as they orbit the earth.  But there are 24 of these planes, which means that each plane is spaced 15° apart along the equator.

All spaceflight is done in terms of the "delta-v" budget, which is simply the amount of change in velocity you need to accomplish the mission.  Most SpaceX flights take 9200 m/s of delta-v to reach a minimum orbit.  That orbit is usually 200 km x 28.6°, which is the latitude of Cape Canaveral, the easiest inclination to get to.

I've taken a swipe at computing how much delta-v each of the orbits in the Starlink constellation will take, using an approximation.  To the 9200 m/s for the standard orbit, I add the difference in orbital speed between the 200 km orbit (7788 m/s) and the altitude of the target orbit.  This is probably a bit conservative, but it's hard to do this stuff without actually simulating the launch, which I'm not willing to do for all the configurations we have here.  Then, to get an idea of the delta-v needed to go to a higher inclination, I take the difference in the earth's rotational speed at Canaveral (408 m/s) and the rotational speed we'd have if we launched at the same latitude as the target orbit.  Again, this is only approximate.

In addition to the set of orbits, we need to know the delta-v required, and what the maximum payload is for each of the launchers in reusable mode.  Finally, once we know that, we can compute the number of Starlinks we can fly, limited by the volume restrictions if necessary.

Here's the full set of orbits for the satellites:

OrbitDelta-v CanaveralF9 PayloadFH PayloadPhaseSatellites in OrbitF9 BirdsFH Birds
550 x 53° orbit9,52717,60036,000115843062
1110 x 53.8° orbit9,82215,70030,800116002753
1130 x 74° orbit9,97814,70029,00014002550
1275 x 81° orbit10,10314,00027,60013752448
1325 x 70° orbit10,04014,30028,20014502449
345.6 x 53° orbit9,41318,40036,200225473262
340.8 x 48° orbit9,37918,70036,700224783262
335.9 x 42° orbit9,34119,00037,200224933362

Note that the very high inclinations will probably be launched from Vandenberg.  I should have re-biased them for the different launch latitude, but it was too much work.  It's close enough.

The number of satellites per launch isn't anything close to generating an integral multiple of the orbital plane.  This would appear to be bad news, because you can't just launch a rocket to one plane, then flit to another one without spending a lot of delta-v.

But there is a trick that allows you to deploy the satellites, then let them hop for free to different planes, as long as the planes are at the same inclination.  Because Earth isn't perfectly spherical, the bulge at the equator has the effect of causing an orbit to "precess", drifting slowly to the west each orbit.  The lower the orbit, the more precession there is.

To get your satellites to take advantage of this, you deploy them a bit lower than their target orbit.  This causes them to precess faster than the target orbital plane, which allows them to slowly move into position in a different plane.  When they get there, they use a small amount of their propellant to boost up to the proper altitude.  From there, all the satellites in the plane will precess at the same rate, so they'll form a fairly stable ring of birds in their orbital plane.

Above, we saw that the planes for the 550 x 53° orbits were spaced 15° apart at the equator.  Suppose that, instead of launching straight to 550 km altitude, we deploy things at 400 km.  In the 400 km orbit, the orbital plane will precess to the west 0.357° per day faster than at the 550 km altitude.  As a result, a Starlink can move from one plane to another in about 42 days.  When it gets there, it'll take only 84 m/s of delta-v to move to the 550 km orbit.

Finally, note also that I don't have any numbers for BFR in here.  That's because... well, we'll get to that.  But first, we need to talk about...


When aircraft fly, they're controlled by a highly mature infrastructure, consisting of airport operations, ground control of the runways and surrounding airspace, and finally by the national air traffic control system.  That infrastructure didn't magically appear overnight; it's the product of more than 100 years of operational experience, and the management of tens of millions of flights.  But now it works, and it's reasonably efficient.

Rockets aren't nearly as mature.  There have only been about 6100 suborbital and orbital launches in human history.  The operations needed to ensure safe launches are complex and poorly standardized. Most important, rockets are thousands of times less reliable than aircraft.  They can and do fall out the sky, zip off in unintended directions, and blow up, either in the air or on the ground.

The best concept we've come up with so far to control these things is called a range.  There are several ranges in the US, but the two we're interested in are the Eastern Range, through which launches from Kennedy Space Center and Canaveral Air Force Station fly, and the Western Range, through which launches from Vandenberg Air Force Base fly.

The range is responsible for, among other things:
  1. Managing communications to and from the vehicle.
  2. Monitoring and communicating weather information.
  3. Tracking of the launch.
  4. Clearing out airspace prior to a launch.
  5. Moving the public and ground operations personnel out of the area around a launch.
  6. Clearing maritime traffic from under the path of the launch.
  7. Coordinating recovery efforts.
  8. Ensuring that errant launches are destroyed before they can damage stuff on the ground.
Above all, the range operations are there to ensure the safety of the public.  NASA and the Air Force consider a launch to be safe if there is less than a 1 in 10,000 chance of hurting somebody on the ground under any possible track that the launch can take.

This is a lot like air traffic control, but it's vastly more primitive.  It's more like air traffic control if you expected one out of every thirty or forty 737s that took off to crash into a subdivision at the end of the runway.

Right now, before a rocket can launch, the range has to confirm that the airspace along the trajectory is clear for hundreds of kilometers.  It has to do the same for maritime traffic.  It requires days to weeks of coordination with other agencies to accomplish this.  To make things even worse, launches are often delayed for hours waiting for upper level winds to die down, or for a last-minute glitch to be debugged.  Launches are horribly disruptive to the air and sea space around them.

In 2018, the Eastern Range will likely handle 20 or 21 launches this year, and they have a "drive to 48" campaign that's suppose to get the to 48 launches per year by 2024.  The Western Range will only handle 9 or 10, and its capacity will grow very slowly.  Range capacity is by far the limiting factor on how many launches can occur.

Launch Azimuth

Each range also has a set of headings that it can accommodate.  These are a function of the geography, and are chosen to keep the rockets and the stuff that falls off of them from landing on people.  The Eastern Range can handle launch azimuths from 35° to 120° (due north is an azimuth of 0°).  The Western Range can handle azimuths from 158° to 201°.  The East is therefore good for launching things into fairly low orbital inclinations between 28° and 57°, while the West is good for polar orbits, between 70° and 104°.  (An inclination greater than 90° is a retrograde orbit, where the orbital path rotates in the opposite direction from the way the Earth rotates.  Since launches get a lot of free delta-v from the rotational speed of the Earth, retrograde orbits are hard to reach.)

Now we're ready to talk about BFR.

What About BFR?

When it comes to Starlink, BFR is... well, complicated.

One the one hand, one of the justifications for Starlink is to provide a revenue stream for BFR development, so using BFR to launch Starlink has a certain chicken-vs-egg problem.  However, a partial Starlink constellation can start to provide revenue long before the full constellation is in place, so maybe BFR, with its high capacity, is what allows Starlink to be deployed before the FCC's drop-dead dates.

Another obvious problem is schedule risk.  When you absolutely have to have almost 6000 satellites on-station by late 2024, it would be real bad if, instead of going operational in late 2022, BFR was delayed by a couple of years.  And since we're talking about something being scheduled using Elon's personal multiplier, that's a definite possibility.

But when you actually schedule things out against the existing and planned capacity in the two ranges from which SpaceX operates, it soon becomes apparent that it's very, very difficult to get enough birds on-orbit to meet the deadlines without BFR.

In addition to its huge capacity, BFR comes with a brand new range, one that's operated by SpaceX out of Boca Chica, Texas, on the coast right next to the US-Mexico border.  There's no special magic that SpaceX possesses that can make range operations super-efficient, but an extra 12-24 launch slots a year makes a big difference.  That number of launches, with the raw capacity of the BFR, makes hitting the Starlink numbers pretty easy.

But... uh, there's a problem.  A pretty big one.

Remember that 1 in 10,000 chance of injuring someone on the ground?  That implies that for BFR to be adequately safe, it needs to stay out over the water until the BFS is close to achieving orbital velocity.  I did a rough simulation of a BFR/BFS launch and, while the BFR core booster has separated and headed back to the launch site for reuse within about 150 km, the BFS doesn't make orbit until it's almost 1500 km downrange.

If you take a look at a map, you'll notice that it's pretty hard to stay out over the water when launching from Boca Chica.  There's really only a narrow range of azimuths that send the BFS through the Florida Straits, south of the Bahamas.  Fortunately, that's exactly where you want to go if you're launching heavy payloads to GEO or interplanetary orbits.

But it's not where you launch Starlink satellites.

For every orbital inclination, there's both a northerly and southerly launch azimuth.  Here they are for the Starlink orbits from Boca Chica:

OrbitNorth Azimuth from BCSouth Azimuth from BC
550 x 53 orbit42.0138.0
1110 x 53.8 orbit41.1138.9
1130 x 74 orbit17.9162.1
1275 x 81 orbit10.0170.0
1325 x 70 orbit22.4157.6
345.6 x 53 orbit42.0138.0
340.8 x 48 orbit48.1131.9
335.9 x 42 orbit55.8124.2

That's a pretty dry set of numbers, so let's plot out some of the interesting ones (leaving out the really high inclinations, which would launch out of Vandenberg) on the map, out to 1500 km:

The problem is pretty obvious:  The trajectory for the interesting Starlink orbits goes over millions of people.  Now, the actual computation of the risk to the public requires looking at every possible point of the path, estimating the chance of something falling on that spot, and then biasing that chance by the nearby population density.  So I can't completely rule out some of the southerly paths that go over a lot of Yucatan jungle, but my intuition tells me that it doesn't look good.

But there is another possibility:  Instead of launching direct to the proper inclination, the BFR could launch through the Florida Straits, achieve orbit, and then execute what's called a "plane change maneuver", where we change the inclination from orbit.

The problem with these is that they're hideously expensive in terms of delta-v.  In fact, they're so expensive that even a single BFS with no payload can't perform both the launch and the needed plane change with its total available propellant.

However, the BFS is designed to be refuelable on-orbit.  So if we launched a bunch of Starlinks, then launched two additional BFS tankers to refuel the one with the Starlinks, we have enough delta-v for some of the inclinations.

Here's the delta-v, payload, and number of Starlinks chart for BFR, as launched from Boca Chica into a 26° inclination orbit, refueled twice, and then executing a plane change to the target orbit:

OrbitDelta-v to 26 degree orbit from BCBFR Plane Change Delta-vBFR Payload With 2 Refuels and Plane Change From BCBFR Birds
550 x 53 orbit9,5183,54326,70015
1110 x 53.8 orbit9,8123,50721,90012
1130 x 74 orbit9,9685,93000
1275 x 81 orbit10,0936,66800
1325 x 70 orbit10,0315,39200
345.6 x 53 orbit9,4033,59726,60015
340.8 x 48 orbit9,3692,94158,90034
335.9 x 42 orbit9,3322,146100,00057

The number of Starlinks for each launch is horrifyingly low for a rocket that's designed to launch 100 tonnes into LEO.  That's because that number is biased by the fact that it takes 3 launches to get to the target orbit at all.  The numbers listed are an average covering all three launches.  For example, the BFS actually takes 45 birds to the 500 x 53° orbit, but we divide by 3 to come up with the average number.

So BFR is not a great platform for Starlink.  However, it does have two things going for it:
  1. The per-launch cost is being advertised as very cheap.  ($10M gets bandied about a lot, but I doubt even SpaceX knows yet.) 
  2. The BFR can launch from that lovely, uncrowded new range.  So if we're range-limited elsewhere (and we are), throwing an extra 20 or 30 launches per year at the problem helps a lot.
It should be noted, however, that even if the BFR is pretty cheap, the low average number of birds per launch gives the Falcon Heavy a cheaper launch cost per Starlink for all orbits except the 340.8 x 48° and the 335.9 x 42° orbits.  However, cost per satellite really doesn't matter if you can't get what you need launched in time.  The extra range capacity provided by Boca Chica is worth the premium SpaceX will pay for using the BFR.

So What's the Answer?

I can't say I'm completely happy with the model I put together, but I compiled some data on range and pad resources, along with the non-SpaceX demands on them, the SpaceX non-Starlink demands, and finally some semi-reasonable scheduling for the Starlink demands.  All of these come with an initial yearly launch rate, an annual growth rate, a maximum rate beyond which they can't grow, and a date at which operations can start.

Here's what I used:

Range CapacityDate AvailableResources When First AvailableAnnual Resource GrowthMax Per YearRangePadLauncher
Pad Capacity
Non-SpaceX Demand
ULA Western Range1-Jan-19110%6WestOtherA5/Vulcan
ULA Eastern Range1-Jan-19510%10EastOtherA5/Vulcan
ULA BC Range1-Jan-9900%0TexasOther-
Blue Origin Western Range1-Jan-9900%0WestOther-
Blue Origin Eastern Range1-Jan-20425%8EastOtherNew Glenn
Blue Origin BC Range1-Jan-9900%0TexasOther-
OmegA Western Range1-Jan-21110%3WestOtherOmegA
OmegA Eastern Range1-Jan-21210%4EastOtherOmegA
OmegA BC Range1-Jan-9900%0TexasOther-
Non-Starlink SpaceX Demand
Non-Starlink F9 West SLC-4E1-Jan-19510%10WestSLC-4EF9
Non-Starlink F9 East LC-401-Jan-19135%20EastLC-40F9
Non-Starlink F9 (crewed) LC-39A1-Jan-1925%3EastLC-39AF9/D2
Non-Starlink FH LC-39A1-Jan-19210%4EastLC-39AFH
Non-Starlink BFR BC1-Jan-21020%20TexasBC-
Starlink Demand
550 x 53 orbit1-Jul-19210%5EastLC-39AFH
1110 x 53.8 orbit1-Sep-20210%5EastLC-39AFH
345.6 x 53 orbit1-Jul-201210%100EastLC-40F9
1130 x 74 orbit1-Jul-19410%100WestSLC-4EF9
1275 x 81 orbit1-Jul-20410%100WestSLC-4EF9
1325 x 70 orbit1-Jan-21310%100WestSLC-4EF9
345.6 x 53 orbit1-Jul-201210%100EastLC-39AFH
340.8 x 48 orbit1-Jul-201210%100EastLC-39AF9
340.8 x 48 orbit1-Oct-221225%100TexasBCBFR
335.9 x 42 orbit1-Oct-221225%100TexasBCBFR
550 x 53 orbit1-Jul-19410%4EastLC-40F9
1110 x 53.8 orbit1-Sep-20510%5EastLC-40F9
1130 x 74 orbit1-Jul-191210%100WestSLC-4EF9
1275 x 81 orbit1-Jul-201210%100WestSLC-4EF9
1325 x 70 orbit1-Jan-211210%100WestSLC-4EF9
345.6 x 53 orbit1-Jul-201210%100EastLC-39AFH
345.6 x 53 orbit1-Jul-201210%100EastLC-40F9
340.8 x 48 orbit1-Oct-221225%100TexasBCBFR
335.9 x 42 orbit1-Oct-221225%100TexasBCBFR

With this, I got:

  • 2210 phase 1 Starlinks by the end of 1Q2024.
  • 3830 phase 2 Starlinks by the end of 4Q2024 (which actually isn't quite good enough, but it's literally close enough for government work.
Both of these numbers meet the FCC's half-constellation "use it or lose it" rule.

The big takeaway, however, is that I simply can't get this work without two things:

  1. Starlinks starting to launch from the Eastern Range on both FH and F9 platforms by the third quarter of 2019--about 7 months from now.
  2. The BFR launching at least once a month starting in 4th quarter of 2022, and growing by 25% a year thereafter.  
So the success or failure of Starlink comes down to executing pretty crisply on the BFR, and getting the Texas range humming along at a pretty good clip as soon as the BFR is ready to go.

Things would get a lot easier if the Eastern Range wasn't just driving to 48, but instead went to something like 56-60 launches a year.  But my guess is that that isn't going to happen.  As a result, a lot of the burden of executing on this will fall on Boca Chica and the BFR.  (Remember:  birds go up on the BFR only every third launch; the other two are refueling runs, which likely require that SpaceX has a fleet of at least 6 BFRs to get started.)

Things also rapidly get out of control if SpaceX can't start launching Starlinks in high volume some time in the next 7 or 8 months.  That's not a lot of time.

What About Regulatory Relief?

SpaceX has repeatedly asked the FCC for a relaxation of the the "use it or lose it" rule, given that coming even close will demonstrate their earnestness in using the licensed spectrum, and also given that the range restrictions aren't really going to be their fault.  So far, the FCC has denied these requests, but have invited SpaceX to apply for the waiver at a later date.

So it's not completely implausible that the FCC might take pity on them if things are going pretty well, but they can't quite drum up the capacity soon enough.

On the other hand, SpaceX as a company and Elon as a person have made some powerful enemies, and Starlink poses an existential threat to an awful lot of vested interests:

  1. One of the best ways of killing the BFR is to kill the funding needed to complete it, and Starlink revenue is an integral part of that funding. Boeing, Lockheed-Martin, and Northrop-Grumman are all painfully aware that the success of the BFR could easily take them out of the launch business.  When you hear the phrase, "military-industrial complex", these guys ought to be in your mental picture every time.  They have something like 70 years of experience lobbying legislators and regulators in Washington.  SpaceX is a babe in the woods compared to them.
  2. The telecom industry is another heavily entrenched, politically sophisticated industry that stands to lose big from Starlink, although this is a mixed bag:  Starlink would be a definitive, permanent answer to the mobile telecom industry's need for backhaul bandwidth (the communications links that take stuff from the cell towers and dump it to the terrestrial network, and vice-versa).  But it's also kind of a nightmare, because Starlink has the ability to grow to be a formidable competitor to incumbent mobile networks around the world.
  3. Finally, there are a whole bunch of people in Washington who cordially hate Elon Musk.  They don't like his flashiness.  They don't like his politics.  They don't like the fact that he makes no bones about wanting to squeeze the life out of the world's fossil fuel infrastructure as quickly as possible.  None of these people would piss on him if he were on fire, to say nothing of going to bat for him with the FCC. 
The bottom line here is that SpaceX would be crazy to count on regulatory relief.  Yes, they might get it, but there's an excellent chance that they won't.  They'll be planning accordingly.


The bottom line is that Starlink is doable, but it's really, really, really tight.  Even minor hiccups in either the manufacturing of the satellites or SpaceX launch operations could cause them to miss the crucial milestones.  Significant delays in the BFR can mess things up.  A serious launch accident can mess things up.

I'm not quite sure where the crunch will come for Elon and his group.  It could be early next year, if the Starlink satellite development isn't proceeding smoothly.  By 2020, SpaceX will need to be making Starlinks at a rate of about 600 a year.  By 2022, they'll need to be making 1500 a year.

The crunch could be with BFR, or with the Boca Chica range.  Even fairly modest delays put the phase 2 deadlines in serious jeopardy.

Or it could just be the sheer, grinding, sustained operational tempo that's required to pull this off.  Make no mistake:  Starlink will require SpaceX to go flat-out, launching as many missions as they possibly can, for at least the next 10 years.  Launch crews will get worn down.  Key people will quit.  It's going to be a challenge to maintain an organization that's robust enough to deal with the inevitable hiccups and still keep going, satellite after satellite, launch after launch.

It's going to be hellish.  But one thing's for sure:  If SpaceX can pull this off, the world will be changed forever.  I'm rooting for them.  But I'm worried that failure could pose an existential threat to SpaceX as a company.

Update 12/17/2018:

One of the things that hadn't occurred to me is that SpaceX can launch the BFR through the Yucatan Channel and get to a 33.2° inclination.  That then makes the plane changes about 1000 m/s cheaper, which is huge.

I've reworked the number of birds per platform, using this idea.  I've also figured out exactly how much prop is needed to reach the target inclination, and assumed that a BFS tanker (aka a BFS with no payload and some extra prop in its tanks) can loiter waiting for the next launch of Starlinks, somewhat reducing the number that have to be launched.  I've assumed a 20% boiloff while loitering--that number is derived purely by Rectal Extraction.

Here's what I get:

OrbitDelta-v For F9, FH Canaveral (Also Vandenberg but it's somewhat wrong)Delta-v Boca Chica to 33.2 InclinationBFR Plane Change From 33.2 Delta-vF9 PayloadFH PayloadBFR Payload With Refuel and Plane Change From 33.2BFR Refuelings Needed For Plane ChangeBFR 20% Boiloff Adjusted Launches Per Load of Starlinks
550 x 53 orbit9,5279,5082,61017,60036,00088,3002.63.72
1110 x 53.8 orbit9,8229,8032,61015,70030,80069,80034
1130 x 74 orbit9,9789,9595,08214,700#N/A#N/A#N/A#N/A
1275 x 81 orbit10,10310,0845,85014,000#N/A#N/A#N/A#N/A
1325 x 70 orbit10,04010,0214,54314,300#N/A#N/A#N/A#N/A
345.6 x 53 orbit9,4139,3942,64918,40036,20095,9002.63.72
340.8 x 48 orbit9,3799,3601,98518,70036,70098,2001.72.84
335.9 x 42 orbit9,3419,3221,18319,00037,200100,0000.92
OrbitF9 Birds Per LaunchFH Birds Per LaunchBFR Birds Per Cargo + Prop LaunchesF9 Cost Per BirdFH Cost Per BirdBFR Cost Per Bird
550 x 53 orbit306241$396,667$280,645$243,902
1110 x 53.8 orbit275330$440,741$328,302$333,333
1130 x 74 orbit25#N/A#N/A$476,000#N/A#N/A
1275 x 81 orbit24#N/A#N/A$495,833#N/A#N/A
1325 x 70 orbit24#N/A#N/A$495,833#N/A#N/A
345.6 x 53 orbit326244$371,875$280,645$227,273
340.8 x 48 orbit326260$371,875$280,645$166,667
335.9 x 42 orbit336286$360,606$280,645$116,279
F9 CostFH CostBFR Cost

Note that I've also included a per-Starlink launch cost for each of the three platforms.  My cost numbers are derived from an exercise I did a while ago, looking at SpaceX's launch costs for the F9 and FH assuming that cores got reused 10 times.

With these new numbers, I can get both the phase 1 and phase 2 deadlines met, using only 48 Eastern Range launch slots and delaying the BFR until 2Q2023.  However, it requires a pretty aggressive ramp, both on the Eastern and Texas ranges, to get there.

Yet Another Update, 12/19/18:

Fixed the previous table--I had the wrong F9 and FH costs.  BTW, my F9/FH cost model is here.

I've also gone through the scheduling exercise in a bit more detail, which has allowed me to put a total cost on both the manufacturing and launch of the entire constellation.  Assuming it costs $100K to build a Starlink bird, the total cost is about $4.4 billion.



Unknown said...

I also then wonder about the viability of one web and Boeings sat constellations...wouldn't they be under threat as well? Lots of big dreams becoming reality now that viable business models exist

TheRadicalModerate said...

Boeing's constellation is stalled and hasn't yet been approved by the FCC. OneWeb has recently scaled back their constellation from 900 birds to 600 birds. But yes, if OneWeb were to start making heavy use of New Glenn, that would result in still more congestion on the Eastern Range. Note that I've got New Glenn starting out at 4 launches/year in 2020, rising up to no more than 8/year by 2023. That might be too conservative.

Unknown said...

TRM, great write up. I wonder if SpaceX might seek launches at other facilities, for example the European, Japanese or New Zealand ranges. Doing so might take some pressure off the Texas range and BFR plans. Of course there would be added costs to use them. Your thoughts?

TheRadicalModerate said...

The problem with going elsewhere is that SpaceX still has to build a pad. The other problem is that all SpaceX launchers are restricted by ITAR, the regulations that control export of technologies deemed to have national security implications.

I could see them going after a pad on Wallops, but I doubt that a range sitting right next to the mouth of the Chesapeake is going to have the kind of capacity they need. There have been some speculations about doing something in Puerto Rico, but there's no infrastructure to speak of--it would be like spinning up Boca Chica all over again, but with terrible logistics.

John Bucknell said...

TRM. Great article. I like your financial analyses, but I wonder if you might take a deeper dive into BFS/BFR costs to NEOs. I find the number of $1000/kg to lunar surface challenging to meet due to the number of on-orbit refuelling missions it will take to achieve such a trip - even one way.

TheRadicalModerate said...


To be honest, I'm not sure where the BFR number I cited came from.

I did some analysis a while back on the Adelaide version of the BFR/BFS, in terms of getting it to the lunar and surface and back with no refueling. I think I eyeballed the results and dropped a decimal--they're quite a bit higher than the $1000 I stated. Assuming that the cargo launch refuels first in LEO, then also in HEEO (+2500 m/s delta-v), from a tanker that it itself has been refueled in LEO before rendezvousing in HEEO, and assuming $10M per launch, I get the following:

15 launches total: 10,000 kg/launch ($15,000/kg)
14: 9800 kg/launch ($14,300/kg)
13: 8500 kg/launch ($15,300/kg)
12: 7100 kg/launch ($16,900/kg)
9: Minimum possible, at only 700 kg/launch ($128,600/kg)

Note that this would be considerably worse with the DearMoon version of the BFR/BFS. But they're massively better if you have refueling capability on the lunar surface, or even in LLO.

I inferred the FH mass to the surface from the Zubrin Moon Direct phase 1 numbers, and again my math is a bit off, although not by nearly as much: I get $14,900/kg, instead of $12,000/kg.

But these are not apples-to-apples comparisons, since Zubrin's looking at an expendable lander and we should assume that BFRs are to expensive to expend. Zubrin also has some numbers for the BFR, but they assume a methalox lander deployed in LEO or LLO. Those come out to 20.1 t and 82.7, respectively. I frankly don't understand these: I get a number closer to 40 t deploying from LEO, and I can't get LLO to work at all without refueling. (Perhaps he's assuming an expendable BFR, which ain't gonna happen...) However, note that 40 t from LEO to the surface comes out to only $250/kg with one $10M BFR launch.

So you have a lot of things to choose from here. By the far the best case is to build a big, honkin', expendable, methalox lander for the BFR and deploy it from LEO, but there may be some optimum that's better than this.

Anyway, thanks for the note. I'll make a note in the text that what I have is wrong, or at least needs more work.

John Bucknell said...

Thanks, this is much closer to what I was expecting.

It is also worth noting the the $10M per launch cost of the BFR/BFS assumes a LOT of reuses of both vehicles. The Space Review article on the 2017 model assumes $350M for one each of BFR/BFS (