Author Topic: Starlink : General Discussion - Thread 1  (Read 1217312 times)

Offline pagheca

  • Bayesian Pundit. Maybe.
  • Full Member
  • ****
  • Posts: 759
  • Lives in Ivory, Tower
  • Liked: 220
  • Likes Given: 161
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #500 on: 02/10/2015 07:23 am »
Without the specifics of the project is difficult to say, but with 4,000 satellites moving on the sky there would be on average 1 every 130,000 km2. Let's do a tenth by optimization of orbital parameters to improve coverage of low-latitude areas. Still 13,000 km2. NYC area is, for comparison, 1,214 km2.

So, you may have just a bunch of satellites covering NYC (several of them may overlap). Free to think this is comparable the way ground based internet works (more nodes => more connections where needed) as a stand-alone, user-to-user solution, but I will not buy it without further confirmations.
« Last Edit: 02/10/2015 08:02 am by pagheca »

Offline Nomadd

  • Senior Member
  • *****
  • Posts: 8839
  • Lower 48
  • Liked: 60430
  • Likes Given: 1305
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #501 on: 02/10/2015 08:23 am »
Most of those sats will be idling while flying over inhabited areas, while a few will be struggling for a few minutes with zillions of high-demanding users.

I can't see a sat internet network siding ground based competitors. I would rather see sats network backing up fibers and radio ones in semi-deserted areas. That would make price very high.
Those sats over pacific/atlantic are not idling, but serve as the backbone/transocean bridge for the "over land" sats. And if you think about how many paths via different sets of sats you can have over pacific (assuming the direct links can be established between sats orbiting not adjacent to each other), you realize Elon's desire to compete with transatlantic fiber...
If they do go high speed laser or such between sats, this could make the undersea cable operators pretty nervous.
« Last Edit: 02/10/2015 08:23 am by Nomadd »
Those who danced were thought to be quite insane by those who couldn't hear the music.

Offline ChrisWilson68

  • Senior Member
  • *****
  • Posts: 5266
  • Sunnyvale, CA
  • Liked: 4992
  • Likes Given: 6459
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #502 on: 02/10/2015 09:05 am »
It seems like a few relatively simple steps can improve tolerance quite a bit. For example choosing a SOI process for the chip (which many fabs use anyway) seems to help. The avionics would want to do more than the communications payload.

[...]

4000 satellites with multiple chips is getting to the sort of scale where an ASIC makes sense, this is plausible. You can build multiple cores onto a single chip and have them do their votes on the chip. Still need multiple chips but I suspect the part count can be reduced compared to the current Dragon system.

I think you are greatly underestimating the cost of doing an ASIC comparable to off-the shelf multi-processor DSP chips.  An enormous amount of engineering goes into optimizing a chip like that.  4000 is a tiny, tiny volume for an ASIC.  And it's not just the cost (maybe $100 million to design an ASIC like that), it's the time it takes.  If they have to build a chip design team from scratch, don't expect the first production chips for 3 years.  There's also the risk -- they might not find out until 2 years in that their chip won't have the performance they projected.  Meanwhile, TI comes out with a new, improved multi-chip DSP that beats the ASIC.

The chips are tiny and very cheap.  This is exactly the kind of thing where SpaceX can leverage its cheap launch costs -- just use cheap off-the-shelf chips and pay for the extra mass it takes to have lots of redundant chips, shielding, or both.

Offline speedevil

  • Senior Member
  • *****
  • Posts: 4406
  • Fife
  • Liked: 2762
  • Likes Given: 3369
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #503 on: 02/11/2015 11:07 am »
The chips are tiny and very cheap.  This is exactly the kind of thing where SpaceX can leverage its cheap launch costs -- just use cheap off-the-shelf chips and pay for the extra mass it takes to have lots of redundant chips, shielding, or both.

Also, in principle, because most of the time it's not over a major metropolitan area, you could have 90% of the chips 'idle' - and using no power.
You only burst it to 20* nominal power when you really, really need it.
The power and thermal requirements are somewhat easier if you only need to do this once every 5 orbits (say) for 8 minutes.

Offline JamesH

  • Full Member
  • ****
  • Posts: 525
  • United Kingdom
  • Liked: 284
  • Likes Given: 7
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #504 on: 02/11/2015 12:51 pm »
It seems like a few relatively simple steps can improve tolerance quite a bit. For example choosing a SOI process for the chip (which many fabs use anyway) seems to help. The avionics would want to do more than the communications payload.

[...]

4000 satellites with multiple chips is getting to the sort of scale where an ASIC makes sense, this is plausible. You can build multiple cores onto a single chip and have them do their votes on the chip. Still need multiple chips but I suspect the part count can be reduced compared to the current Dragon system.

I think you are greatly underestimating the cost of doing an ASIC comparable to off-the shelf multi-processor DSP chips.  An enormous amount of engineering goes into optimizing a chip like that.  4000 is a tiny, tiny volume for an ASIC.  And it's not just the cost (maybe $100 million to design an ASIC like that), it's the time it takes.  If they have to build a chip design team from scratch, don't expect the first production chips for 3 years.  There's also the risk -- they might not find out until 2 years in that their chip won't have the performance they projected.  Meanwhile, TI comes out with a new, improved multi-chip DSP that beats the ASIC.

The chips are tiny and very cheap.  This is exactly the kind of thing where SpaceX can leverage its cheap launch costs -- just use cheap off-the-shelf chips and pay for the extra mass it takes to have lots of redundant chips, shielding, or both.

Whilst the timescales are about right, you can do an ASIC for a lot less than that. Tapeout costs about $1M depending on process, so you got to add on design,layout, testing, software etc. I'm working in a team that does this sort of stuff (Although I'm software side), there are 7 of us, so not huge in labour costs. The chip we are doing will cost probably $3M to develop. And will be quite complex...but we never start from scratch, there are always libraries you bring in to save time. As an existing example, the Brcm2836, the chip in the new Raspberry Pi, took about 1.5yrs to develop, with a team of about 10. However, that reused a lot of work from the previous chip.

So the cost does depend on how much reuse you can have, and therefore how much new work is required, and how much of it there is to do. But $100M is much too high, probably double.

And, tbh, you would do this in an FPGA anyway, an ASIC would be much more expensive for the low volumes involved.

Offline Robotbeat

  • Senior Member
  • *****
  • Posts: 39270
  • Minnesota
  • Liked: 25240
  • Likes Given: 12115
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #505 on: 02/11/2015 01:15 pm »
BTW, I really doubt they'll develop their own ASIC.

But the volume will be much higher than 4000. They'll have at a very minimum 3 times that due to their redundancy/reliability strategy, and if you add in multiple chips to handle the workload, you easily could be talking over a dozen chips per satellite. Dragon uses dozens of chips, for instance.

So you could be talking 50k-100k chips (though likely of various types) for each constellation iteration.

But it's true that they'd want to use state of the art (non-space-sector-specialized) chips in order to stay on the cutting edge without having to invest a lot every year.

But where a custom ASIC (customized from a licensed existing ASIC design, most likely) would make sense is for the ground terminals, since there would be millions of them, perhaps eventually hundreds of millions.

But they'll need to invest in custom phased array technology (on both ends) which may itself be chip-scale. But given the number of elements needed (hundreds of thousands or even a million per satellite), they'd easily have enough volume to justify custom chips.
Chris  Whoever loves correction loves knowledge, but he who hates reproof is stupid.

To the maximum extent practicable, the Federal Government shall plan missions to accommodate the space transportation services capabilities of United States commercial providers. US law http://goo.gl/YZYNt0

Offline JamesH

  • Full Member
  • ****
  • Posts: 525
  • United Kingdom
  • Liked: 284
  • Likes Given: 7
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #506 on: 02/11/2015 03:03 pm »
I would think they would need make in excess of a million to make the decision to go with an ASIC, prior to that any recent FPGA would do (assuming they are robust enough).

Offline ArbitraryConstant

  • Senior Member
  • *****
  • Posts: 2014
  • Liked: 628
  • Likes Given: 311
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #507 on: 02/11/2015 03:27 pm »
I think you are greatly underestimating the cost of doing an ASIC comparable to off-the shelf multi-processor DSP chips.
Well, there's two main areas. The avionics and the communications payload. Both have potential advantages to doing custom chips.

For the avionics it's mostly a matter of shrinking the overhead necessary for redundant voting computers, so it would be mostly pretty simple processors, licensable stuff. I thought about whether the layout and design rules might be different for radiation tolerance, but even if they are the actual computation power needed for this isn't huge, so simple synthesizable cores are viable options, so most of the changes here would happen via automated tools. This isn't the same as a highly optimized design.

It's different for the communications payload, but that's also a lot more forgiving of errors so COTS makes more sense. If customization were desired here, again, a lot of it is probably licensable. The customization might even mean going to TI and asking if they can do a custom version of a DSP with the same cores but slightly different features, like maybe spare cores for failures that don't happen on the ground, and going outside the normally economical die size envelope.

So the cost does depend on how much reuse you can have, and therefore how much new work is required, and how much of it there is to do. But $100M is much too high, probably double.
For the avionics it seems like it would be quite simple ARM cores and some SRAM and some communications to implement the redundant/voting architecture Dragon uses with less mass, power, and without a pressurized compartment.

At a few hundred kg I'd expect most of that to go to power, propulsion, antenna, with most of the rest going towards the communications payload. If a relatively easy ASIC can improve radiation tolerance while reducing a significant circuit board to 1-2 chips that may be worth it. Obviously subject to a better analysis of the trades than I can provide. But I do think this is going to have to optimize mass more aggressively than Dragon does.

Offline Robotbeat

  • Senior Member
  • *****
  • Posts: 39270
  • Minnesota
  • Liked: 25240
  • Likes Given: 12115
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #508 on: 02/11/2015 11:02 pm »
I would think they would need make in excess of a million to make the decision to go with an ASIC, prior to that any recent FPGA would do (assuming they are robust enough).
Considering there probably isn't a chip scale phased array solution that meets their needs, that seems like a fairly easy number for them to hit (not that the development effort will be easy!). They're going to need billions of phased array elements, total, for the constellation and billions more for the ground terminals.
Chris  Whoever loves correction loves knowledge, but he who hates reproof is stupid.

To the maximum extent practicable, the Federal Government shall plan missions to accommodate the space transportation services capabilities of United States commercial providers. US law http://goo.gl/YZYNt0

Offline ArbitraryConstant

  • Senior Member
  • *****
  • Posts: 2014
  • Liked: 628
  • Likes Given: 311
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #509 on: 02/11/2015 11:26 pm »
But the volume will be much higher than 4000. They'll have at a very minimum 3 times that due to their redundancy/reliability strategy, and if you add in multiple chips to handle the workload, you easily could be talking over a dozen chips per satellite. Dragon uses dozens of chips, for instance.
What made me think it might be worth it it is that Dragon has an embarrassment of excess performance and a handy pressure vessel to put everything in, and the computers have much less work to do. A satellite as mass sensitive as what's been proposed may want to shrink this.

Could mean limited customization of existing IP blocks or special orders from existing companies rather than doing something from whole cloth.

Offline JamesH

  • Full Member
  • ****
  • Posts: 525
  • United Kingdom
  • Liked: 284
  • Likes Given: 7
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #510 on: 02/12/2015 12:07 pm »
I would think they would need make in excess of a million to make the decision to go with an ASIC, prior to that any recent FPGA would do (assuming they are robust enough).
Considering there probably isn't a chip scale phased array solution that meets their needs, that seems like a fairly easy number for them to hit (not that the development effort will be easy!). They're going to need billions of phased array elements, total, for the constellation and billions more for the ground terminals.

Slightly different from the FPGA/ASIC stuff though - this is the transmission side rather than the processing side.

Offline Robotbeat

  • Senior Member
  • *****
  • Posts: 39270
  • Minnesota
  • Liked: 25240
  • Likes Given: 12115
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #511 on: 02/12/2015 02:08 pm »
I would think they would need make in excess of a million to make the decision to go with an ASIC, prior to that any recent FPGA would do (assuming they are robust enough).
Considering there probably isn't a chip scale phased array solution that meets their needs, that seems like a fairly easy number for them to hit (not that the development effort will be easy!). They're going to need billions of phased array elements, total, for the constellation and billions more for the ground terminals.

Slightly different from the FPGA/ASIC stuff though - this is the transmission side rather than the processing side.
Well yeah, you can't use an FPGA for that. Has to be ASIC.
Chris  Whoever loves correction loves knowledge, but he who hates reproof is stupid.

To the maximum extent practicable, the Federal Government shall plan missions to accommodate the space transportation services capabilities of United States commercial providers. US law http://goo.gl/YZYNt0

Offline gospacex

  • Senior Member
  • *****
  • Posts: 3024
  • Liked: 543
  • Likes Given: 604
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #512 on: 02/12/2015 02:28 pm »
There is one thing I am positively certain about. They won't be using commercially available space grade arrays unless the producers find a way to sell them a huge lot cheaper than they do now.

What is the price?

Offline Robotbeat

  • Senior Member
  • *****
  • Posts: 39270
  • Minnesota
  • Liked: 25240
  • Likes Given: 12115
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #513 on: 02/13/2015 02:30 am »
There is one thing I am positively certain about. They won't be using commercially available space grade arrays unless the producers find a way to sell them a huge lot cheaper than they do now.

What is the price?
I think it's >>$100/Watt, which would be a large portion of the satellite costs if using those traditional cells. Dragon's arrays are most certainly using cheaper cells than traditional.
Chris  Whoever loves correction loves knowledge, but he who hates reproof is stupid.

To the maximum extent practicable, the Federal Government shall plan missions to accommodate the space transportation services capabilities of United States commercial providers. US law http://goo.gl/YZYNt0

Offline BobHk

  • Full Member
  • ***
  • Posts: 324
  • Texas
  • Liked: 91
  • Likes Given: 173
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #514 on: 02/14/2015 02:05 pm »
Going into the future a bit does anyone expect non solar power to be used on the sats around Mars?  It seems a perfect time to use an ASRG rather than Solar.

Offline AncientU

  • Senior Member
  • *****
  • Posts: 6257
  • Liked: 4164
  • Likes Given: 6078
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #515 on: 02/14/2015 02:26 pm »
Going into the future a bit does anyone expect non solar power to be used on the sats around Mars?  It seems a perfect time to use an ASRG rather than Solar.

Europa Clipper is being baselined with solar power... Jupiter is at 5.2AU.  Mars is at 1.5AU.
Inverse square law... Mars solar = 12x Jupiter solar.
« Last Edit: 02/14/2015 02:27 pm by AncientU »
"If we shared everything [we are working on] people would think we are insane!"
-- SpaceX friend of mlindner

Offline BobHk

  • Full Member
  • ***
  • Posts: 324
  • Texas
  • Liked: 91
  • Likes Given: 173
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #516 on: 02/14/2015 02:32 pm »
Going into the future a bit does anyone expect non solar power to be used on the sats around Mars?  It seems a perfect time to use an ASRG rather than Solar.

Europa Clipper is being baselined with solar power... Jupiter is at 5.2AU.  Mars is at 1.5AU.
Inverse square law... Mars solar = 12x Jupiter solar.

Which was a cost decision....old space.  18 square meter solar panels are a bit hard to repair on something around Mars if you want them to work for years eh?



Offline guckyfan

  • Senior Member
  • *****
  • Posts: 7438
  • Germany
  • Liked: 2332
  • Likes Given: 2891
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #517 on: 02/14/2015 02:46 pm »

18 square meter solar panels are a bit hard to repair on something around Mars if you want them to work for years eh?

The likelihood that something goes wrong with  a mechanical system over many years is orders of maginitude higher than for solar. Also solar would have likely at least two panels which leaves you with still half power if something goes wrong with one.

Offline ArbitraryConstant

  • Senior Member
  • *****
  • Posts: 2014
  • Liked: 628
  • Likes Given: 311
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #518 on: 02/14/2015 09:50 pm »
Which was a cost decision....
If this is meant to suggest SpaceX won't make cost decisions for Mars satellites I think you're wrong.

18 square meter solar panels are a bit hard to repair on something around Mars if you want them to work for years eh?
No harder than Juno's 24 m2 panels around Jupiter, or any of the current Mars orbiters.
« Last Edit: 02/14/2015 09:57 pm by ArbitraryConstant »

Offline BobHk

  • Full Member
  • ***
  • Posts: 324
  • Texas
  • Liked: 91
  • Likes Given: 173
Re: SpaceX - now a satellite manufacturer (Starlink)
« Reply #519 on: 02/15/2015 11:28 pm »
Which was a cost decision....
If this is meant to suggest SpaceX won't make cost decisions for Mars satellites I think you're wrong.

18 square meter solar panels are a bit hard to repair on something around Mars if you want them to work for years eh?
No harder than Juno's 24 m2 panels around Jupiter, or any of the current Mars orbiters.

ESA and NASA are constrained by budget.  if Elon decides he'd rather have a 100 year power plant (Amerecium 241 has a half life od 400 years; in a stirling engine/rtg based power solution 100 years is not so hard)  in a Mars orbiting sat rather than solar panels he has the $ to do so, the ESA was not so lucky: they sought and were DENIED the budget to go RTG/ARTG.

edited: for spelling 
« Last Edit: 02/15/2015 11:34 pm by BobHk »

Tags:
 

Advertisement NovaTech
Advertisement Northrop Grumman
Advertisement
Advertisement Margaritaville Beach Resort South Padre Island
Advertisement Brady Kenniston
Advertisement NextSpaceflight
Advertisement Nathan Barker Photography
0