SpaceX Internet Satellite Network: Starlink

[Cosmacelf]

Interesting article about how to radiation harden chips for use in space. I wonder how Starlink is doing this? I would think the chips needed for their gigabit packet processing would be pretty specialized meaning they would be looking to designing their own chips. Hmmm.

Space-grade CPUs: How do you send more computing power into space?

 

[traxila]

Am eagerly awaiting the opportunity to invest.

Forge

 

[BioSehnsucht]

Interesting article about how to radiation harden chips for use in space. I wonder how Starlink is doing this? I would think the chips needed for their gigabit packet processing would be pretty specialized meaning they would be looking to designing their own chips. Hmmm.

Space-grade CPUs: How do you send more computing power into space?

Probably the same way SpaceX does for everything else they send to space - i.e., they won't do anything special. Especially since Starlink satellites are expected to have short lifetimes and be easily replaced.

Generally speaking they use off the shelf cheap mass produced processors (whether they be ARM or x86 or some other flavor) and instead of having one or two "radiation hardened" processors (which generally speaking are slow and ancient because the finer the lithography the more likely a random strike will break something, and thus you can't really radiation harden more modern tech) they have greater redundancies by way of more cheaper systems (which even with mass penalties and additional hardware comes out far cheaper than buying radiation hardened processors).

 

[mongo]

Out of investor main

I think the pentagon is likely to be the first client, even if just a 100 million pilot program. The DOD will want to know how encryption is handled on the chip and in memory. I think they’ll do better then a carrier solution and be NIST compliant. This is huge for all the branches. I wouldn’t be surprised if SpaceX and the military find a way to sneak a few hundred observation devices in with the regular devices.
The consumer market is easy and huge. For a global company though, this would be awesome. Providing a single network provider. Sell this as a backbone to Azure and companies like Zoom and Ring Central. This could be the AWS of networking. If Elon wanted to truly give some deep deep love to the shorts he’d combine Spacex and Tesla once starlink is up and monetized at 100 billion. He’d own ~40% of the combined companies and have insane cash flow. I’ve never thought a merger would happen and probably not going to happen, but SpaceX is going to be very valuable and I’d like to own some shares.

Military would not put any traffic on a public network without encrypting it first. So any SpaceX encryption is only interesting to them from an eavesdropping PoV.

SpaceX will be low profit (and remain private) as they pour all their money into Mars colonization. Elon said a while ago maybe once they stop growing (dividend zone) then maybe go public.

 

[Cosmacelf]

Probably the same way SpaceX does for everything else they send to space - i.e., they won't do anything special. Especially since Starlink satellites are expected to have short lifetimes and be easily replaced.

Generally speaking they use off the shelf cheap mass produced processors (whether they be ARM or x86 or some other flavor) and instead of having one or two "radiation hardened" processors (which generally speaking are slow and ancient because the finer the lithography the more likely a random strike will break something, and thus you can't really radiation harden more modern tech) they have greater redundancies by way of more cheaper systems (which even with mass penalties and additional hardware comes out far cheaper than buying radiation hardened processors).

Did you read the article? The Russians recently tried to use off the shelf chips and ended up destroying their mission. A telecom satellite has to work with very high uptime. It can't be rebooting all the time. Something has to be done to mitigate random bit flips. If the system design is to duplicate all processing paths, well, wouldn't you need best of three voting? Anyways, presumably SpaceX has figured this out since they do have craft that stay in space for reasonable periods of time (Dragon).

 

[jerry33]

Did you read the article? The Russians recently tried to use off the shelf chips and ended up destroying their mission. A telecom satellite has to work with very high uptime. It can't be rebooting all the time. Something has to be done to mitigate random bit flips. If the system design is to duplicate all processing paths, well, wouldn't you need best of three voting? Anyways, presumably SpaceX has figured this out since they do have craft that stay in space for reasonable periods of time (Dragon).

You can either harden the chips or harden the environment where the chips live. I imagine that it's less expensive to do the environment (but not necessarily easier). Hardening the environment means you can switch chips as better ones become available. It also will help to use server class chips rather than consumer class chips (yes, there is a difference).

 

[mongo]

Did you read the article? The Russians recently tried to use off the shelf chips and ended up destroying their mission. A telecom satellite has to work with very high uptime. It can't be rebooting all the time. Something has to be done to mitigate random bit flips. If the system design is to duplicate all processing paths, well, wouldn't you need best of three voting? Anyways, presumably SpaceX has figured this out since they do have craft that stay in space for reasonable periods of time (Dragon).

A telecom satellite does, a mega constellation with multiple sats with multiple antennas that can see the same site and only stay pair with it for minutes can be less robust. RAID 5 for satellites, if you will.
Random bit slips are something critical (rocket and automotive) software already deal with. Single redundancy is fine if you checksum to filter the failure (worst case, TCP/IP will do it for you) and can just pull the channel/ sat out of service on hard failure, reboot/ reload on soft.

Back in the day, vendor literature compared FPGAs in airborne applications due to higher incidence of bit flips.

 

[bxr140]

...satellite laser links. Since that's a core part of the system, its interesting they don't have an info graphic on that.

That’s because they don’t exist yet. Its difficult tech to get down to the imperative price point for their application.

If you take a look at this launch’s stack, you’ll see each sat has what appears to be two ~30 cm dishes mounted on two axis gimbals. These are almost certainly RF ISLs, which is an interesting development. They had to relocate the GPS antennas*** [relative to the first launch a few months ago] way out on brackets, ostensibly for field of view clearance, which suggests these ISLs could be a bit of an afterthought. Two is also an interestingly small number for ISLs, but does allow for communication to one in-plane and one adjacent-plane satellite which is kind of all you really need, so…smart. Don’t launch more than you need.

***The GPS antennas are painted black on this launch, and are mounted on long, black, flat brackets. Also super clever, the solution to structural support for those long brackets is just some printed some white spacers that basically form a single vertical column all the way up the stack. Those spacers also provide support for the sort of clunky square bracket that’s under (or over in the photo) the dishes. I can’t tell if those square features support an additional phased array or not.

 

[bxr140]

Interesting article about how to radiation harden chips for use in space. I wonder how Starlink is doing this?

Likely the same way they approach every problem: Take the best of the what industry has done to date, layer in first principals, and—imho most important—iterate margins with empirical data. Fold in the element of volume and it’s a recipe for low cost and high reliability. It really is that self-evident; the legacy industry really is that backward.

COTS has been the hold grail of industry, but its hard to close the case on just dealing with a bunch of SEEs and general degradation. Shielding can help, but there's a limit where it doesn't make a difference (thickness asymptote depends on your shielding material, of course). Its actually easy to find rad tolerant parts for most of what's on your electronics board--sometimes space grade parts are literally best-lots or lots that have simply gone through additional screening. Sometimes a COTS part is actually statistically more reliable than a Space variant that is otherwise identical, because the COTS parts are built at such high volume that you naturally have more consistent yield, even if manufactured on the same exact machine. For more complicated parts, like ICs, one plausible path is to buy a gazillion COTS items and then individually screen them to pull out the best performing parts that are better simply as a function of normal variation. I don't think SpaceX would do that in this application because of the volume, but...who knows.

Long story short, their ICs are definitely going to be some degree of rad tolerant and are going to have some degree of shielding, pretty much like everything else in space, but implemented in a way that's like nothing else in space.

 

[BioSehnsucht]

Did you read the article? The Russians recently tried to use off the shelf chips and ended up destroying their mission. A telecom satellite has to work with very high uptime. It can't be rebooting all the time. Something has to be done to mitigate random bit flips. If the system design is to duplicate all processing paths, well, wouldn't you need best of three voting? Anyways, presumably SpaceX has figured this out since they do have craft that stay in space for reasonable periods of time (Dragon).

It's been known (to those that follow SpaceX closely) that they don't use radiation hardened gear but instead cheap off the shelf components with increased redundancy for their existing vehicles, so there's no reason to expect anything different.

What Hardware/Software Does SpaceX Use To Power Its Rockets? - RankRed

Rather than using expensive, radiation-hardened components, SpaceX uses off-the-shelf parts. According to the former director of SpaceX vehicle certification, John Muratore, each Dragon spacecraft is equipped with 3 flight computers. Each of these computers run on a dual core x86 processor.

The systems do not utilize multicore capabilities of a processor. Instead, they perform each computation on the 2 cores separately and compare the results. Therefore, 3 flight computers with dual core processor act as 6 independent computers that are regularly verifying each others calculations.

If one of the flight computers outputs different value (due to radiation), the others will detect it. In this case, the malfunctioning processor is automatically rebooted to prevent further errors. It copies the memory from other processors and executes the same programs to get up to speed with what other systems are executing. This is called re-sync.

...

The Falcon 9 rocket is packed with 3 flight computers for each engine, and triple redundancy computer, which overall carries 30 processors. We are presenting 2012 data, so it is possible that SpaceX is using even more processors in their spacecraft and vehicle to handle the landing.

...

This is what they call a radiation-tolerant design, and it’s much different than a radiation-hardened design. Even NASA does not use radiation hardened components everywhere. For instance, ISS (International Space Station) uses a mix of both radiation-hardened and radiation-tolerant components, along with conventional laptops for some basic controls.

(bolding not mine, but from the article)

 

[Cosmacelf]

Anyone have a clue who is going to manufacture their ground based antenna terminals? Unlikely SpaceX since it is a high volume consumer product. Who makes phased array antennas these days anyways?

 

[BioSehnsucht]

Anyone have a clue who is going to manufacture their ground based antenna terminals? Unlikely SpaceX since it is a high volume consumer product. Who makes phased array antennas these days anyways?

I believe these will basically be just large PCBs with components on them, and the exterior will be pretty ordinary too, so there should be no shortage of available places to contract out the construction of the user terminals to.

That said, PCBs aren't exactly rocket science and neither are pick and place machines or reflow ovens, or injection molding, so depending on how many plan to make at what rate (and whether perhaps it could be a joint venture with Tesla - build PCBs for Tesla, make use of Tesla's plastic facilities for housings, etc) it's not impossible they vertically integrate it. The actual SMT components and such will come from the usual suppliers of course, but the actual manufacturing isn't outside the realm of mostly-vertical integration.

That said, I would expect the most likely scenario would be to sub it out to a Taiwanese or Chinese firm, the former more likely for political reasons usually especially since the DoD is interested, but Elon's on good terms with China so who knows - they could always just license the design to DoD for them to use their own subcontractors to build DoD terminals. They could also sub it out to someone based in the USA but there is almost zero high volume PCB companies that actually operate in the US these days, and you generally pay quite a premium for the patriotism of it.

 

[e-FTW]

I believe these will basically be just large PCBs with components on them, and the exterior will be pretty ordinary too, so there should be no shortage of available places to contract out the construction of the user terminals to.

That said, PCBs aren't exactly rocket science and neither are pick and place machines or reflow ovens, or injection molding, so depending on how many plan to make at what rate (and whether perhaps it could be a joint venture with Tesla - build PCBs for Tesla, make use of Tesla's plastic facilities for housings, etc) it's not impossible they vertically integrate it. The actual SMT components and such will come from the usual suppliers of course, but the actual manufacturing isn't outside the realm of mostly-vertical integration.

That said, I would expect the most likely scenario would be to sub it out to a Taiwanese or Chinese firm, the former more likely for political reasons usually especially since the DoD is interested, but Elon's on good terms with China so who knows - they could always just license the design to DoD for them to use their own subcontractors to build DoD terminals. They could also sub it out to someone based in the USA but there is almost zero high volume PCB companies that actually operate in the US these days, and you generally pay quite a premium for the patriotism of it.

I want to say Elon has said before that they would make those themselves.

 

[BioSehnsucht]

I want to say Elon has said before that they would make those themselves.

Finally assembly themselves isn't that big of a deal, but making the PCBs or housings is a bit more involved, so unless they plan to joint venture that with Tesla for greater economies of scale (as they might do for a foundry for Cybertruck / Starship steel) it seems like an odd move to insource everything.

 

[mongo]

Launches planned for December and January.
SpaceX announces second Starlink satellite launch in two weeks

 

[e-FTW]

SpaceX working with the American Astronomical Society:
SpaceX to experiment with less-reflective satellite coatings on next Starlink launch – Spaceflight Now

“The goal of Starlink is to provide worldwide internet service, an aspiration we do not want to impede, but this requires one to two orders of magnitude more low Earth orbiting satellites than currently exist,” Krafton wrote. “We do not want to give up access to optical observations from the ground. Our group’s task is to find a path forward that accommodates both uses of the sky.”

 

[Ben W]

SpaceX working with the American Astronomical Society:
SpaceX to experiment with less-reflective satellite coatings on next Starlink launch – Spaceflight Now

Vantablack?

In seriousness, there are a several different approaches to brightness mitigation. One is see if the constellation can fly lower, where it would be visible for a shorter time after sunset and before sunrise. The lowest Starlink orbit is about 340km (210mi), which will typically be brightly visible (magnitude +4 to +6) about an hour after sunset and an hour before sunrise. Higher orbits are actually worse; the satellites are dimmer but can stay visible much of the night, and there will appear to be many more of them in the sky, and they will appear to move slower, causing brighter streaks on astronomical exposures.

Another is to see if Starlink's goal can be achieved with fewer satellites. (12000 - 42000 is a LOT of satellites, especially if other companies/nations start making copycat constellations.)

Another, of course, is to apply dark coatings. The downside is that this significantly heats up the satellite, increasing radiation in the infrared (among other side effects), which may greatly interfere with IR astronomy even if the satellite is dark at optical wavelengths.

Another is to design the spacecraft to reflect light upward/outward rather than downward. A Mylar "cone of shame" (pointy end up) with the spacecraft recessed within the cone would geometrically accomplish this, while also shielding the spacecraft from sunlight. The catch is that the solar panels still need sunlight, and the panels are likely the source of most of the apparent brightness, not the body of the spacecraft.

So another is to tilt the solar panels to reflect sunlight upward, during the problematic twilight period. This would reduce power to the satellite while incurring its own power demands, as well as cause wear and tear on the moving parts. But it could be done only on demand when passing specific astronomical sites, e.g. Mauna Kea.

It will be interesting to see what solutions SpaceX comes up with. I do hope Elon and Co. take it seriously.

 

[Buckminster]

Vantablack?

In seriousness, there are a several different approaches to brightness mitigation. One is see if the constellation can fly lower, where it would be visible for a shorter time after sunset and before sunrise. The lowest Starlink orbit is about 340km (210mi), which will typically be brightly visible (magnitude +4 to +6) about an hour after sunset and an hour before sunrise. Higher orbits are actually worse; the satellites are dimmer but can stay visible much of the night, and there will appear to be many more of them in the sky, and they will appear to move slower, causing brighter streaks on astronomical exposures.

Another is to see if Starlink's goal can be achieved with fewer satellites. (12000 - 42000 is a LOT of satellites, especially if other companies/nations start making copycat constellations.)

Another, of course, is to apply dark coatings. The downside is that this significantly heats up the satellite, increasing radiation in the infrared (among other side effects), which may greatly interfere with IR astronomy even if the satellite is dark at optical wavelengths.

Another is to design the spacecraft to reflect light upward/outward rather than downward. A Mylar "cone of shame" (pointy end up) with the spacecraft recessed within the cone would geometrically accomplish this, while also shielding the spacecraft from sunlight. The catch is that the solar panels still need sunlight, and the panels are likely the source of most of the apparent brightness, not the body of the spacecraft.

So another is to tilt the solar panels to reflect sunlight upward, during the problematic twilight period. This would reduce power to the satellite while incurring its own power demands, as well as cause wear and tear on the moving parts. But it could be done only on demand when passing specific astronomical sites, e.g. Mauna Kea.

It will be interesting to see what solutions SpaceX comes up with. I do hope Elon and Co. take it seriously.

Each satellite has 4 antennae pointing in the same direction - presumably towards earth? I know that these can be pointed in different directions as they are phased array but won't the angle be too acute for satellite to satellite routing?
Is the 90 angle between the solar panel and satellite body fixed?

 

[jerry33]

It will be interesting to see what solutions SpaceX comes up with. I do hope Elon and Co. take it seriously.

Create an affordable orbiting platform for amateur astronomers so they can take their own pictures from space.

 

[e-FTW]

Is anyone considering Starlink for airplanes (aside from the military)?
https://ipv6.org.uk/wp-content/uplo...eeting_2019_IPv6_in_Aviation_SITA_Dec2019.pdf