SpaceX Internet Satellite Network: Starlink

[Bipo]

Thanks! Then they will move to 5° offset between planes. Nice.

I see that they're also thinking about 70° planes as well as SSO. Nice to know how many on those, but since the orbits are so much close on those latitudes, the number of satellites needed to grant service should be much lower.

 

[bxr140]

Will those new satellites still shift into three different planes? Maybe since the full 360° are already covered (in 20° slices) the follow on batches will fill up those planes or even start to complete the polar part of the constellation...

FWIW for a given amount of sats coverage is generally better with more planes and fewer sats in each plane. So...I'd expect ongoing deployment of the constellation to focus more on additional planes rather than more sats in each plane, though both will of course be enhanced.

I see that they're also thinking about 70° planes as well as SSO. Nice to know how many on those, but since the orbits are so much close on those latitudes, the number of satellites needed to grant service should be much lower.

Yeah, and there's so few people in those regions that it doesn't make a lot of sense to spend massively disproportionate effort/capital on that part of the constellation. If 1% (and probably much less) of your total users are serviced by those high inclination shells, 0th order ratio suggests a ~4000 satellite constellation would dedicate ~40 sats to those shells, but reality will be easily 3-5% of the total number of sats, if not more. Ultimately coverage will likely end up with as low as 2 sats in those regions (basically just for the sake of redundancy, but potentially diversity...not sure if SpaceX is going that route...) with ISLs (instead of gateways) providing the backhaul. Compare that to even the initial 1500 sat constellation for places like North America (especially in the high 40's low 50's) that will likely have coverage from 3-5 or even more sats.

SSO makes sense from a rideshare perspective, though its still a little unclear how financially beneficial the rideshare program will be for Starlink so the bottom line is still a bit fuzzy. Its also pretty crappy for augmenting the primary 53° constellation because the sats spend so little time over the populated regions that would need additional/surge coverage. SSO is largely for marketing: "Global coverage".

A 70deg sub constellation (or thereabouts) basically grabs the last couple percent of the world's population that's above 60° and can better augment the primary constellation since the sats spend more time on average over serviceable areas--one could also imagine a scenario where this constellation provides some or all of the air/sea service, and potentially non-continental service as well.

One thing is for sure, the traffic scheduling/routing is going to be extremely complex!

 

[Wahbash]

.Each launch of 60 sats is currently filling 3 planes...

So, I assume each plane is home to each of the different paths weaving their way past each other in the animations? All the sats going in the same direction share the same plane?

 

[Off Shore]

One thing is for sure, the traffic scheduling/routing is going to be extremely complex!

What are your thoughts about applying neural networks to the problem?

 

[bxr140]

So, I assume each plane is home to each of the different paths weaving their way past each other in the animations?

Yes. The orbits are near perfect two dimensional circles in their respective planes; when overlayed on a mercator projection like we see in some of the animations, each orbit appears as a sine wave. The three space animations obviously represent the same thing, though it can be a little hard to visualize each LEO orbit as a two-dimensional plane since the altitude is so low (and so the earth takes up most of the animation.

Here's a random 20 second YouTube illustrating the a single plane:

If you take that single plane and rotate 18 degrees about the earth's spin axis you'll get to the next starlink plane...and so on. And yes, complicating matters is that the earth is also rotating under the planes, so the 18 degrees has to be referenced to something other than what intuitively might be a point on the surface of the earth. The Sun is the typical reference.

Here's a clip of the GPS constellation just for visualization. Its a little easier to see each orbital plane since the altitudes are way higher (so the earth is smaller) and there are many fewer planes (so it is a lot less cluttered).

All the sats going in the same direction share the same plane?

Yes. Its much easier to achieve the necessary ground coverage with fewer sats when they're all going the same direction in a plane. In the animations you can see the ground coverage circles within each plane move in lock step, since the sats within a plane are all stationary relative to one another. So from a modeling perspective each plane's coverage is more or less just a band as opposed to a bunch of circles.

There are other reasons you wouldn't want them going opposite directions: Obviously collisions are a concern, but they could be avoided by having the altitude of the forward orbit 'lane' ~10-20km or so different than the reverse. Launch efficiency is an issue too--its much more energy efficient to launch in the direction of the earth's rotation, which means you can launch more mass then if you were launching 'backward'. Its a play on spitting into the wind, more or less... Last is enabling the intersatellite link--there's almost no way you'd get opposite moving sats to close any kind of link between them when they're in close range since the closing velocity would be crazy.

 

[bxr140]

What are your thoughts about applying neural networks to the problem?

I'm definitely not the right person to provide a detailed analysis, but certainly that's a logical progression. There's a number of potentially limiting factors in the satellite network that realistically can only be optimized and re-optimized as the constellation evolves, and presumably many of those factors/variables are at best ill-defined right now, let alone not even known.

Many of those variables are pretty straightforward and really are just a matter of processing data. As a case study, consider long haul flights: As the data set grows, so will the ability to correlate Starlink system utilization to things like actual departure and actual flight times, flight plans and weather deviations, number of PAX, etc. That could very helpful for Starlink capacity planning, since the sats will largely use time over the oceans to cool off thermally and recharge batteries in support of the significantly higher load over populated regions. If weather is going to bump all the trans-Atlantic flights way north, for instance, that may be near the fringe of constellation coverage where a single sat is going to be loaded more than, say, a way-south flight path that might put all the flights (At least from the east coast) right in that low-50's sweet spot of max coverage.

What gets interesting is when you dig into the secondary and tertiary variables where some more neural tool could excel. In that same case study, how far in advance can things like airfield and flight path weather predictions estimate departure time? Maybe number of PAX/bookings can predict boarding delays? Maybe that heavy departing at the same time from the next gate only flies 3 times a week, and the resulting crowd in the terminal slows down boarding? Can a social media scrape identify a butterfly delay (Twitter A-hole: "WTF my plane is broken!!1!!1!") that will ripple around the world and could be far better at identifying delays than anything else?

Or maybe all of that is a human induced wild goose chase that would be discarded in milliseconds by Hal?

Anyway, you get the point. So, yeah, bring on the machine.

 

[adiggs]

What are your thoughts about applying neural networks to the problem?

This sounds like more of an optimization problem to me, than a neural network problem. Neural net type problems are more like supervised learning - use historical data along with historical outcomes to find patterns that will enable you to predict future outcomes using current data.

Orbital mechanics / equations are known (even if I don't know them). The only thing trick here is that you have a LOT of moving parts that need optimization. New object to put into the mix, when and how much each satellite needs to be raised (and it'd be nice if they didn't run into each other), etc..

Optimization problems are characterized by having 1 or more metrics / criteria (measurable) that are being optimized for. Here, maybe that's time to final orbit, minimal risk of any collisions, minimal fuel consumption, ...

I wouldn't want to set this up, but that's mostly because my knowledge of optimization techniques is minimal, where my knowledge of supervised learning has me employed in the field.

 

[e-FTW]

https://twitter.com/tramblis/status/1278071875243925504?s=20

 

[dhanson865]

another site for visualizing starlink orbits

Stuff in Space

gives results like this that you can spin around

 

[Buckminster]

The competition:
Setting up a private satellite Wi-Fi network is now cheaper and easier than ever | ZDNet
Low bandwidth seems to be the biggest issue.

 

[Cosmacelf]

The competition:
Setting up a private satellite Wi-Fi network is now cheaper and easier than ever | ZDNet
Low bandwidth seems to be the biggest issue.

It’s dial up modem slow. 72 kbps. A 2 Mbps link for today’s internet is slow, I can’t imagine how painful 72 kbps would be.

 

[bxr140]

The competition:
Setting up a private satellite Wi-Fi network is now cheaper and easier than ever | ZDNet
Low bandwidth seems to be the biggest issue.

You may want to read past the headline. Nobody--including the article--would ever suggest Iridium or Inmarsat or Globalstar be an access solution for anyone but folks in the least connected regions who need the most basic of connectivity and who are willing to pay a premium for the trouble. Broad brushing them together, their primary customer base is people that need mobile service in remote areas.

For funsies, check out the service plans here: Satellite Phone & Internet Service Plans & Deals - Satellite Phone Store

 

[doghousePVD]

The competition:
Setting up a private satellite Wi-Fi network is now cheaper and easier than ever | ZDNet
Low bandwidth seems to be the biggest issue.

This is just a regurgitated press release. No specs, reviews or comments, just fluff. Almost unusable for the vast majority of users due to radio shadows from trees, buildings, hills, smokestacks on a cruise ship, etc. (I had an Iridium phone for several years, very disappointed.)

 

[TheJaguar]

CosmaCelf

It’s dial up modem slow. 72 kbps. A 2 Mbps link for today’s internet is slow, I can’t imagine how painful 72 kbps would be.

It isn't too bad. I was able to sign up for this website, verify through my email and post this comment using a software limited 72 kbps web browser.

 

[Cosmacelf]

I just got an email from Starlink.com asking for my address instead of just zip code. The email said “Starlink private beta begins this summer with public beta to follow”.

 

[bkp_duke]

I just got an email from Starlink.com asking for my address instead of just zip code. The email said “Starlink private beta begins this summer with public beta to follow”.

Same, although given where we live I don't have high hopes to be in the public beta.

 

[ecarfan]

I just got an email from Starlink.com asking for my address instead of just zip code. The email said “Starlink private beta begins this summer with public beta to follow”.

As did I. Since I am located in an urban area I consider my chance of getting Starlink to be essentially zero but I signed up anyway. No way to tell them that I want it for my RV and I am a perfect test subject.

 

[dhanson865]

"Welcome to Starlink Beta

Thank you for participating in our Friends and Family Beta Testing program! This invitation can only be applied to the service address listed above. Please review our Frequently Asked Questions about our beta program before proceeding.

To participate as a beta tester, you will need a clear view of the northern sky from wherever you plan to install your Starlink dish (roof or ground). If you do not have a clear view of the northern sky, please email

[email protected]

These charges are not a fee for the Starlink hardware or services, but are being requested exclusively to allow for the testing of our ordering and billing systems as part of this beta program. SpaceX is temporarily loaning you the hardware and providing the internet services free of charge. The $1 will be charged 30 days after your hardware is shipped. This invitation is not transferable to any other address. By clicking the above link you are activating Starlink Services and authorize regularly scheduled charges to the payment method on file."

Install guide will be at https://www.starlink.com/assets/documents/Install Guide.pdf

mounting types https://preview.redd.it/dajubkl6ota...bp&s=04319e97636f580ab425de0d45c7bbc33038cb6f and https://preview.redd.it/kd8d3i8rota...bp&s=72f17e79bf402e2d56e2ced66b8bcd97a3e5d94e

 

[dhanson865]

I find the part about a view of the northern sky a bit odd unless the beta is strictly for southern California or some specific neighborhood that has mountains to the south.

https://twitter.com/elonmusk/status/1283142384268779522

Starlink terminal has motors to self-orient for optimal view angle. No expert installer required. Just plug in & give it a clear view of the sky. Can be in garden, on roof, table, pretty much anywhere, so long as it has a wide view of the sky.

Yeah, will take less than a minute to order on http://Starlink.com when it goes live

 

[dhanson865]

How starlink is going to be cheaper than the old school sat internet

Total mass launched to orbit in the second quarter of 2020, by rocket builder (Source: Bryce Space and Technology):

1) SpaceX

2) CASC

3) Mitsubishi Heavy Industries

4) Roscosmos

5) United Launch Alliance