Starlink : Markets and Marketing

[RedLineTrain]

Here's an interview of a former ground segment manager at SpaceX who seems to have been part of the Hawthorne takeover of the Seattle office.  Sounds like he was there for several years.  Some good detail, but the manager was careful to stick to Elon-disclosed information.  He was interviewed by a consultant in the cable industry.

One detail.  I think that the manager was wrong when he stated that SpaceX will try to extend the lives of the satellites as much as possible.  Instead, in my opinion, SpaceX will try to turn to new technology levels as quickly as possible.  Satellites and launches will be cheap.

https://app.tegus.co/guest/view/8Y247zVsJbuTv1SNpoNEtquSWTRdCUZLi9ccDiVYtob6kmvJc9sXrnrtUbF9

A teaser...

Former Manager of Starlink Gateways at SpaceX
Yes. So, when I joined in 2016, it was being run and if I'm recalling correctly. So, a lot of this information has been out there. But it was being directed by a team that actually moved over to Amazon Kuiper, which is the Amazon competitor to SpaceX.
And essentially, Elon had said this publicly, but he just kind of wasn't happy with the technical direction that, that team was going down. They were not producing a satellite system, not an individual satellite but a satellite system that scaled at the desired unit economic and performance levels.
So ultimately, it was not feasing to technical leadership and maybe project execution speed. So, a little bit slow. In summary, it's probably a little bit too slow and a little bit off the desired technical excellence route that Elon sets for his managers.

Tegus Client
Yes. Just internationally, I guess there's international emerging markets where I'm super bullish Starlink and then international kind of developed markets, a Germany or a Britain or something. And I would kind of think they look similar to how I'm picturing the U.S. where Starlink's a massive competitor in rural areas and maybe not as much competitor domestic and suburban. Am I thinking about that correctly? Or is there anything unique in international markets that might make them more or less a competitor in those areas?

Former Manager of Starlink Gateways at SpaceX
No. I think any sort of like First World countries is going to look almost identical to the U.S.

[JayWee]

They were not producing a satellite system, not an individual satellite but a satellite system that scaled at the desired unit economic and performance levels.

What does it mean?

[SpaceCadet1980]

Starlink sats are made such that digital packets can be routed/switched from any up/down/ISL links and any other. It is a fundamental requirement to make the system work as defined in the FCC docs. It is not a bent pipe.

can I ask you to quote how it is formulated by the FСС  ?? or give me link to this doc?

I am not sure what FCC document was being referred to, but I am sure you could find it in some "introduction" or "summary" section if you looked. This is just basic fundamental facts about the described architecture, ISLs would be dead weight without it, and direct user to user routing would be impossible. This question seems to be coming from your repeated attempts to describe Starlink as if it was restricted to the architectures used by other systems (such as GEO sat based internet, which is a fundamentally different problem, or OneWeb, which did not seem to be planning on direct to consumer sales, last I checked.)

[vsatman]

This is just basic fundamental facts about the described architecture, ISLs would be dead weight without it, and direct user to user routing would be impossible.

I don't see the role of ISL.
For Satellites v.0.9, a signal at a frequency of 14 GHz came from the Gateway to the satellite , in Ku band down converter  was converted into a 12 GHz signal and sent to the user terminal

Satellites v.1.0 28 GHz signal comes from the Gateway to the satellite in Ka-Ku down converter and  converts into a 12 MHz signal and sent to the user terminal.

Satellites with ISL will work like this, a signal at a frequency of 28 GHz came from the Gateway to the satellite and is converted in RF over fiber Convertor into a signal of 30 (conditionally) THz and sent via ISL to another satellite, there it is again  converted in another RF over fiber Convertor into a signal at a frequency of 12 GHz and transmitted to the terminal.

Why is routing needed here? At the same time, the conversion of 28 GHz to 30 THz, in my opinion, is much simpler, because you can simply modulate a signal in GHz use signal  in THz as carrier.

[JayWee]

Packet routing/switching is absolutely necessary for ISLs to work.
How would you know otherwise where to send the data across permanently-shifting multi-hop mesh network?
How would you implement your simple RF shifting when multiple sats with ISLs are needed? terminal->sat->isl->sat->isl->sat->isl->sat->groundstation
How would you know what goes where?
How would you prioritise packets from premium customers?

You route the packets. That's how.

All you need is a "simple" FPGA.

[Asteroza]

They were not producing a satellite system, not an individual satellite but a satellite system that scaled at the desired unit economic and performance levels.

What does it mean?

They were designing too old school, likely cribbing too much from Teledesic powerpoints, and not innovating as much as Elon desired. If that team wants to go slower, going to a Bezos organization fits better with that ideology. Kicking them out forced a big reset, and resulted in allegedly a number of major improvements (allegedly the switch from a OneWeb-ish bus to the flatpack sat design happened at this time)

[DanClemmensen]

Packet routing/switching is absolutely necessary for ISLs to work.
How would you know otherwise where to send the data across permanently-shifting multi-hop mesh network?
How would you implement your simple RF shifting when multiple sats with ISLs are needed? terminal->sat->isl->sat->isl->sat->isl->sat->groundstation
How would you know what goes where?
How would you prioritise packets from premium customers?

You route the packets. That's how.

All you need is a "simple" FPGA.

Router-in-the-sky is a conceptually simple way to use ISL, and it is what I would choose, but it's not the only possibility. A single optical link (fiber or free-space) can carry hundreds of discrete lambdas (wavelengths), OADMs (optical add-drop multiplexers) can pick off an individual wavelength ("drop") or inject an individual wavelength ("add"), while an optical amplifier can receive and boost and entire set of wavelengths. Optical switches can "circuit-switch" wavelengths from one channel to another. In combination, you can use a separate optical wavelength for each of multiple satellites and relay the wavelengths entirely in the optical domain.  A constantly-changing mesh requires constant routing updates, but has basically the same complexity in the optical circuit domain as in the packet domain. Each satellite must perform the optical-to-RF and RF-to-optical transponder function for its assigned wavelength in the analog domain, and only enough CODEC for its control channel. You need a separate set of satellites for gateway interfaces to convert multiple wavelengths into gateway RF links unless you can figure out how to get optical signals from the ground to/from your gateway satellites.

[JayWee]

https://twitter.com/elonmusk/status/1493358044989767683

Over 250k Starlink user terminals

nice progress!

https://twitter.com/elonmusk/status/1493359783608467459

Yes, but more ground stations & improved packet routing will make a bigger difference

So packet routing it is.

[SpaceCadet1980]

Yes, but more ground stations & improved packet routing will make a bigger difference

So packet routing it is.

He could mean from the Gateway/Ground Station onward… e.g. more PoPs/peering.

There is no question about whether the satellites do packet routing. The basic features of the system including user to user routing require it. Whether this uses traditional switches, or something like DanClemmensen's suggestion about using the carrier frequency of ISLs to indicate the destination isn't particularly relevant.

[DanClemmensen]

Yes, but more ground stations & improved packet routing will make a bigger difference

So packet routing it is.

He could mean from the Gateway/Ground Station onward… e.g. more PoPs/peering.

There is no question about whether the satellites do packet routing. The basic features of the system including user to user routing require it. Whether this uses traditional switches, or something like DanClemmensen's suggestion about using the carrier frequency of ISLs to indicate the destination isn't particularly relevant.

back in my time (2016), the fundamental constraint on a LEO constellation was satellite mass, and required electrical power was the fundamental determinant of mass. This is because solar panel mass and battery mass go up with required power. You need battery because LEO satellites spend slightly less than half their time in the Earth's shadow.

Contrary to the gut feelings of most network engineers, processing bits for data communications takes power, and the more processing steps you take, the more power is consumed. The number of femtojoules per gate flip varies with the type of chips you use. FPGAs are more power-hungry than ASICs. The least amount of bit processing for user data is zero: you get this by transponding in the analog domain with no conversion to digital at all. This is how GEO "bent-pipe" satellites have worked for decades. (Probably) the largest amount of per-bit processing occurs in the demodulation and decoding steps. The more aggressive your MODCOD, the more power you need per bit.

Before I got into the satellite industry in 2004, I spent more than 30 years in terrestrial comms, and I never had to think about power per bit.  I designed and implemented packet switching equipment. As a result, my approach to constellation architecture was to think in terms of packet switching.  We made a proposal for a packet-switched constellation in 2013, but the realities of launch cost forced us to revert to a bent-pipe architecture with no ISL. We then began looking at ISL. The trick of using wavelength-switched ISL completely avoids having any user data in the digital domain in any satellite. It saves a lot of power and therefore a lot of mass. I do not know if anyone ever actually implemented it. The wavelength switching requires at most a few switching operations per second per satellite. Packet switching requires one switching operation per packet, which is millions of switching operations per second, per satellite.

[SpaceCadet1980]

You don't need to lecture me about the importance of a power budget on a satellite. Of course they will want to use any reasonable trick to minimize power use. I must have misread your previous post, because I thought you were still describing an architecture that at least reads enough of the header initially to figure out where to send each packet of data. Skipping the details about how things require power, and cutting to the end where you make your conclusion:

The trick of using wavelength-switched ISL completely avoids having any user data in the digital domain in any satellite. It saves a lot of power and therefore a lot of mass. I do not know if anyone ever actually implemented it. The wavelength switching requires at most a few switching operations per second per satellite. Packet switching requires one switching operation per packet, which is millions of switching operations per second, per satellite.

Not parsing any of the user data at all is simply not a possible answer for Starlink. When data comes in from a user there will be a series of packets, and any given packet may need to go to:
-the local gateway (e.g. to get routed to a local Google data center)
-downlink to a different user (e.g. peer to peer data for a game being played with a friend on Starlink in the next town over.)
-ISL to eventually be downlinked to a gateway or Starlink user on the other side of the country (e.g. video call with family in another state.)

Similar with the data back to the user, it can be coming from multiple places.

Fundamental use case completely kills any attempt to make it no better than a bent pipe and ignoring all of the contents.

They will of course simplify where they can, and it is not like power is free on the ground and excessive processing can't be helpful for latency either. Apparently there are already ways to reduce just how much is parsed by grouping packets:
https://forum.nasaspaceflight.com/index.php?topic=48297.msg2340176#msg2340176
I don't know all of the mechanics of that, but it still needs to at least process a little bit of data first.

[Asteroza]

Yes, but more ground stations & improved packet routing will make a bigger difference

So packet routing it is.

He could mean from the Gateway/Ground Station onward… e.g. more PoPs/peering.

There is no question about whether the satellites do packet routing. The basic features of the system including user to user routing require it. Whether this uses traditional switches, or something like DanClemmensen's suggestion about using the carrier frequency of ISLs to indicate the destination isn't particularly relevant.

back in my time (2016), the fundamental constraint on a LEO constellation was satellite mass, and required electrical power was the fundamental determinant of mass. This is because solar panel mass and battery mass go up with required power. You need battery because LEO satellites spend slightly less than half their time in the Earth's shadow.

Contrary to the gut feelings of most network engineers, processing bits for data communications takes power, and the more processing steps you take, the more power is consumed. The number of femtojoules per gate flip varies with the type of chips you use. FPGAs are more power-hungry than ASICs. The least amount of bit processing for user data is zero: you get this by transponding in the analog domain with no conversion to digital at all. This is how GEO "bent-pipe" satellites have worked for decades. (Probably) the largest amount of per-bit processing occurs in the demodulation and decoding steps. The more aggressive your MODCOD, the more power you need per bit.

Before I got into the satellite industry in 2004, I spent more than 30 years in terrestrial comms, and I never had to think about power per bit.  I designed and implemented packet switching equipment. As a result, my approach to constellation architecture was to think in terms of packet switching.  We made a proposal for a packet-switched constellation in 2013, but the realities of launch cost forced us to revert to a bent-pipe architecture with no ISL. We then began looking at ISL. The trick of using wavelength-switched ISL completely avoids having any user data in the digital domain in any satellite. It saves a lot of power and therefore a lot of mass. I do not know if anyone ever actually implemented it. The wavelength switching requires at most a few switching operations per second per satellite. Packet switching requires one switching operation per packet, which is millions of switching operations per second, per satellite.

If you need a unique wavelength per sat, that starts to get a little hairy when you get into the thousands of sats range? Or would you be potentially mapping to something else, say a specific geographic region or even a global map tile marking a specific destination?

[DanClemmensen]

Yes, but more ground stations & improved packet routing will make a bigger difference

So packet routing it is.

He could mean from the Gateway/Ground Station onward… e.g. more PoPs/peering.

There is no question about whether the satellites do packet routing. The basic features of the system including user to user routing require it. Whether this uses traditional switches, or something like DanClemmensen's suggestion about using the carrier frequency of ISLs to indicate the destination isn't particularly relevant.

back in my time (2016), the fundamental constraint on a LEO constellation was satellite mass, and required electrical power was the fundamental determinant of mass. This is because solar panel mass and battery mass go up with required power. You need battery because LEO satellites spend slightly less than half their time in the Earth's shadow.

Contrary to the gut feelings of most network engineers, processing bits for data communications takes power, and the more processing steps you take, the more power is consumed. The number of femtojoules per gate flip varies with the type of chips you use. FPGAs are more power-hungry than ASICs. The least amount of bit processing for user data is zero: you get this by transponding in the analog domain with no conversion to digital at all. This is how GEO "bent-pipe" satellites have worked for decades. (Probably) the largest amount of per-bit processing occurs in the demodulation and decoding steps. The more aggressive your MODCOD, the more power you need per bit.

Before I got into the satellite industry in 2004, I spent more than 30 years in terrestrial comms, and I never had to think about power per bit.  I designed and implemented packet switching equipment. As a result, my approach to constellation architecture was to think in terms of packet switching.  We made a proposal for a packet-switched constellation in 2013, but the realities of launch cost forced us to revert to a bent-pipe architecture with no ISL. We then began looking at ISL. The trick of using wavelength-switched ISL completely avoids having any user data in the digital domain in any satellite. It saves a lot of power and therefore a lot of mass. I do not know if anyone ever actually implemented it. The wavelength switching requires at most a few switching operations per second per satellite. Packet switching requires one switching operation per packet, which is millions of switching operations per second, per satellite.

If you need a unique wavelength per sat, that starts to get a little hairy when you get into the thousands of sats range? Or would you be potentially mapping to something else, say a specific geographic region or even a global map tile marking a specific destination?

Any given bundle of wavelengths originates at one point and has a limited number of wavelengths. This is similar to the wavelenghts in a single optical fiber. : e.g., 160 lambdas in one optical fiber in the optical C-band, each lambda carrying 10 Ghz For the constellation I was modelling, there were N satellites (N<160) in a plane, so we could hit all of the satellites in a plane using the lambdas in one optical fiber, with one teleport sending the equivalent of all the lambdas to one satellite in the plane. So yes, if you want to hit 16,000 satellites, you would need a teleport link to each of 100 separate satellites. If you have not figured out how to use lasers from a teleport to a satellite, you need some magic, because there is not enough RF bandwidth available, but that's another topic. Note that the optical C-band is optimal for the transparency "sweet spot" in a fiber, near 1500 nm. There is no reason to just use C-band for a free-space optical (FSO)  link, so you are not restricted to those 160 lambdas. Inter-plane links can be added but are harder to explain in a forum post. They are also harder to implement on the satellites because its easier to track your next-ahead and next-behind in your plane than it is to track a satellite in an arbitrary nearby plane.

[JayWee]

Polaris Dawn:

The Polaris Dawn crew will be the first crew to test Starlink laser-based communications in space, providing valuable data for future space communications system necessary for missions to the Moon, Mars and beyond.

So now the ISLs have to be able to support spaceship2starlink too?
Or do they mean radio from Dragon to starlink and then ISL to gateway?

In any case, it further proves that starlink is a router.

[DanClemmensen]

Polaris Dawn:

The Polaris Dawn crew will be the first crew to test Starlink laser-based communications in space, providing valuable data for future space communications system necessary for missions to the Moon, Mars and beyond.

So now the ISLs have to be able to support spaceship2starlink too?
Or do they mean radio from Dragon to starlink and then ISL to gateway?

In any case, it further proves that starlink is a router.

I would guess that they are "routing" in the digital domain also. To be slightly pedantic, the term "router" usually implies IP routing, which they are probably not using. They are probably using a specialized protocol layer with a unique tag per satellite and keeping the forwarding tables for that tags up to date  based on timers that are accurate to the millisecond. Not exactly routing and not exactly tag switching. more like MPLS, but with (I hope) shorter variable-length tags, as in HDLC.

[JayWee]

Yeah, all that Layer 2.5 stuff.

[vsatman]

I would guess that they are "routing" in the digital domain also. To be slightly pedantic, the term "router" usually implies IP routing, which they are probably not using. They are probably using a specialized protocol layer with a unique tag per satellite and keeping the forwarding tables for that tags up to date  based on timers that are accurate to the millisecond. Not exactly routing and not exactly tag switching. more like MPLS, but with (I hope) shorter variable-length tags, as in HDLC.

The architecture of satellite networks is described in the ETSI TR 101 984 (Satellite Earth Stations and Systems (SES);Broadband Satellite Multimedia (BSM);Services and architectures), there are 2 options "transparent network" and "regenerative" with on-board processing - these are their schematic diagrams.

[launchwatcher]

The architecture of satellite networks is described in the ETSI TR 101 984 (Satellite Earth Stations and Systems (SES);Broadband Satellite Multimedia (BSM);Services and architectures), there are 2 options "transparent network" and "regenerative" with on-board processing - these are their schematic diagrams.

The latest version of ETSI TR 101 984 I can find appears to be v1.2.1 from 2007.   I took a quick look and may have missed a few things, but the architectures described don't include any mention of satellite-to-satellite links; all examples I've seen in the document show a single satellite in the datapath between two satellite terminals, and the architecture seems to be wedded to a strict 7-layer model rather than admitting the existence of "layer 2.5" entities like MPLS.

The diagrams in Annex A are odd because they show "peer to peer" IP traffic between the "User ST" and the "Gateway ST" (ST=Satellite Terminal) rather than between devices connected to the "premises" and "external" networks.   Another omission from the architecture diagrams is any mention of the network control center communicating with what the document refers to as the "satellite payload" -- the network elements on board the satellite.   

Figures 6.2, 6.4, and 6.5 are better, showing the difference between mesh and star topologies but to show current possibilities they need to be expanded to include ISL's.   

[su27k]

Elon Musk donates 50 satellite terminals to Tonga

MEIDECC CEO Paula Ma’u expressed his gratitude to SpaceX.

“These terminals will be deployed at strategic locations throughout Tonga to ensure connection and communications are maintained, particularly for the disaster response operation".

Technical staff from SpaceX and the Tonga Government are working on installing the equipment scheduled to be launched next week.

https://twitter.com/mikepuchol/status/1493720085701308423

A little bird tweeted in my ear that two prefixes, 103.152.127.224/27 and 103.152.126.224/27, have been assigned to Nuku'alofa, capital of #Tonga, by @SpaceX #Starlink, under the Sydney POP (confirmed by ping from SYD1). Awesome work @elonmusk & team!

[JayWee]

https://twitter.com/elonmusk/status/1497701484003213317

Starlink sent to Ukraine.