Starlink Internet Connection equipment - Home/Office user

[ZChris13]

1500 elements!

The backside 15 degrees Celsius colder than ambient!

Probably using a peltier as a heat pump to heat the front of the dish. Using a peltier instead of a resistive heater results in about 150% efficiency.

It also lets them cool the front of the dish if a need should arise, right?

[Tommyboy]

1500 elements!

The backside 15 degrees Celsius colder than ambient!

Probably using a peltier as a heat pump to heat the front of the dish. Using a peltier instead of a resistive heater results in about 150% efficiency.

It also lets them cool the front of the dish if a need should arise, right?

Correct, just reverse the polarity and the heat flow will reverse as well.

[LouScheffer]

1500 elements!

This is exactly what you would expect from first principles.  The downlink is 10.7-12.7 GHz, or 2.5 cm wavelength.  The uplink is 14-14.5 GHz, or 2.1 cm.   To avoid hideous sidelobes, you need a spacing of about 0.5 of the shortest wavelength used, or about 1.05 cm. The antenna is 48 cm, allow 2 cm around the rim for margin, so the active diameter is 44 cm.  So you can fit (44/1.05)^2 elements, or roughly 1756 elements. 

[thirtyone]

1500 elements!

The backside 15 degrees Celsius colder than ambient!

Probably using a peltier as a heat pump to heat the front of the dish. Using a peltier instead of a resistive heater results in about 150% efficiency.

Really doubt it, dish is large, thin, no heatsink, and Peltiers are not actually that simple, cheap, or reliable to implement, especially on a curve. For a device of this size it would be an awful waste of money.

Thermal imagers are great at giving really misleading results. Absolute temperature is a measure of both the temperature and surface emissivity of the object. Great for measuring temperature gradients on a uniform surface, not good for absolute measurements. Metallic or reflective surfaces can be notably very, very inaccurate.

I think it's far more likely to be a cold reflection on one side or uncalibrated emissivity for the front vs back of the dish. If someone has a Starlink dish, either just use a thermocouple or use a piece of electrical tape (seriously this is industry standard) on the surface, and take the temperature reading of the tape after it's equilibrated.

[LouScheffer]

I ended up counting a total of 1675 elements in the array.

This highlights the problem of making these things affordable.  If SpaceX wants this antenna to cost less than $1000, the individual ICs must cost less than $0.60 each.  Furthermore, to keep the power consumption down to 100 watts, each can use only 60 mw of power.

These goals are both well beyond the state of the commercially available art.  A standard IC used for receiving this kind of signal, the 4 channel reciever/downconverter ADF5904, cost $26 in volume, and the datasheet shows it consumes 550 mw.

Now this is far from an exact comparison.  On the one hand, the frequency is higher (24 GHz instead of 12) and it has 4 receiving channels where each SpaceX element needs only 2 (one for each polarization).  On the other hand, this is just the low noise amplifier and mixer - each SpaceX element will also need a digitizer and transmitter, plus overall control for summing and phase manipulation.

So SpaceX's only hope, as I see it, is to make this work in a stock commercial CMOS process, and (likely) combine a number of antenna channels (4-16) onto one chip (so one chip for every 8 antennas, perhaps, or about 200 chips).   On the good side, the volumes should be huge.  If they want to build 5,000,000 antennas, with at least 200 chips each, that's an order of 1 billion chips.  That should get the semiconductor manufacturer's attention.

[meberbs]

I ended up counting a total of 1675 elements in the array.

This highlights the problem of making these things affordable.  If SpaceX wants this antenna to cost less than $1000, the individual ICs must cost less than $0.60 each.  Furthermore, to keep the power consumption down to 100 watts, each can use only 60 mw of power.

These goals are both well beyond the state of the commercially available art.  A standard IC used for receiving this kind of signal, the 4 channel reciever/downconverter ADF5904, cost $26 in volume, and the datasheet shows it consumes 550 mw.

Now this is far from an exact comparison.  On the one hand, the frequency is higher (24 GHz instead of 12) and it has 4 receiving channels where each SpaceX element needs only 2 (one for each polarization).  On the other hand, this is just the low noise amplifier and mixer - each SpaceX element will also need a digitizer and transmitter, plus overall control for summing and phase manipulation.

Comparisons like this are not even close to useful for estimating reasonable cost. We don't know what the detection architecture is (though an expert teardown could give come clues.) It isn't actually guaranteed that a up or down conversion would be necessary. More importantly "volume" for that device is on the order of 100s. when Starlink is making thousands of antennas with over a thousand antennas each. That is millions of antenna elements, and they are still at relatively low rate production. These are very different economies of scale and can easily change unit costs by orders of magnitude.

So SpaceX's only hope, as I see it, is to make this work in a stock commercial CMOS process, and (likely) combine a number of antenna channels (4-16) onto one chip (so one chip for every 8 antennas, perhaps, or about 200 chips).   On the good side, the volumes should be huge.  If they want to build 5,000,000 antennas, with at least 200 chips each, that's an order of 1 billion chips.  That should get the semiconductor manufacturer's attention.

This I agree with, with a couple minor caveats. It is pretty clear that they need a custom mass produced chip, and ordering stock parts to build one of these is simply not even an option. I'd expect them to be building this with as much of the RF components printed in a single custom unit. A large complicated IC built for mass production like a CPU can be on the order of ~$100. CPU is obviously very different than the types of circuit elements they would use, but for order of magnitude, this provides some level of plausibility. The other minor caveat is that I think the indications have been that they are building these themselves, the scale they want is certainly at the level to justify buying their own machines to dedicate to the fabrication.

[Asteroza]

Are there low end fabs using old processes using smaller wafers in the 50cm size range? I wonder if it might be cheap enough to blow a whole wafer on per antenna, bugs/defects and all?

[freddo411]

I've have a starlink invite for my new place in MT.    There is a deadline to sign up (Nov 20th for me).    I won't be at my new place for 2 weeks after that.   

Question for the group.   Those of you that sign up and got a dish delivered.   How quickly was the dish delivered after you signed up?

[DreamyPickle]

It would be very interesting to see somebody tear down one of those antennas with a screwdriver. Anyone who does this would probably be kicked out of the beta program but we'd all get to see the PCB.

[LouScheffer]

I ended up counting a total of 1675 elements in the array.

Comparisons like this are not even close to useful for estimating reasonable cost.

Agree this does not help with SpaceX cost.  What is shows is that this is not plausible with existing technology.

We don't know what the detection architecture is (though an expert teardown could give come clues.) It isn't actually guaranteed that a up or down conversion would be necessary.

I'd be quite surprised if there is no down conversion.  Somehow they've got to get the signal into digital form.   Direct sampling would need a sampling rate over 25 Gsamples/sec, technically tough and certainly not within their power budget.  Subsampling still needs a fast sampler (4 Gsamples/sec) and a bandpass filter early in the chain, which is hard to build on an IC.  So I suspect down-conversion which also picks out a sub-band within the overall range.  Anything else takes too much power.  Of course there a chance SpaceX thinks differently than I do.

More importantly "volume" for that device is on the order of 100s.

I don't think this is right - I think the existing volumes are in the millions.   There are at least 3 applications that could support this volume: (1) Military, bulding phased arrays for fighters.  Volume perhaps 500 with 2000 elements each.  (2) Radar based adaptive cruise control for cars.  Volume perhaps a million with 1-2 chips each.  (3) Flat antennas on top of commercial planes for satellite internet.  Volume maybe 10,000 units, with a few hundred elements each.

[...] Starlink is making thousands of antennas with over a thousand antennas each. That is millions of antenna elements, and they are still at relatively low rate production. These are very different economies of scale and can easily change unit costs by orders of magnitude.

The generic learning curve is a 10% cost decrease for a 2x increase in volume.  So a million fold increase in volume (of the same part) would lead to a final cost of 35% of the original.   That's not nearly enough, which is why a custom design is needed, not just cheaper production of existing parts.

The other minor caveat is that I think the indications have been that they are building these themselves, the scale they want is certainly at the level to justify buying their own machines to dedicate to the fabrication.

This I strongly disagree with.  A modern semiconductor fab costs 10s of billions of dollars, and needs thousands of specialized workers.  Their only hope, cost-wise, is to piggyback on someone else's high volume process.  So SpaceX will do the design of the chip, but one of the established semiconductor fabs will do the manufacturing.

[gongora]

What is shows is that this is not plausible with existing technology.

Another way to say that might be "not plausible with existing commercially available components".  I don't think SpaceX really advanced the frontiers of technology here, they put in the work to engineer components to suit their needs since no one else had bothered doing it.

[NekMech]

I've have a starlink invite for my new place in MT.    There is a deadline to sign up (Nov 20th for me).    I won't be at my new place for 2 weeks after that.   

Question for the group.   Those of you that sign up and got a dish delivered.   How quickly was the dish delivered after you signed up?

If I remember correctly, it was something like 6-7 days.
I assume you mean purchase, not sign up.

[meberbs]

I ended up counting a total of 1675 elements in the array.

Comparisons like this are not even close to useful for estimating reasonable cost.

Agree this does not help with SpaceX cost.  What is shows is that this is not plausible with existing technology.

As gongora said, "not plausible with existing technology" is not what you can conclude here. Picking the wrong method of manufacture (buying a ton of COTS parts) only tells you that that method was wrong, which is generally expected.

I'd be quite surprised if there is no down conversion.  Somehow they've got to get the signal into digital form.   Direct sampling would need a sampling rate over 25 Gsamples/sec, technically tough and certainly not within their power budget.  Subsampling still needs a fast sampler (4 Gsamples/sec) and a bandpass filter early in the chain, which is hard to build on an IC.  So I suspect down-conversion which also picks out a sub-band within the overall range.  Anything else takes too much power.  Of course there a chance SpaceX thinks differently than I do.

It is a relatively new trend for more direct RF technology to be available, it can simplify things as mixing always has its own issues. If I had to guess, I'd guess down conversion, but I wouldn't bet on it, because they will have looked at anything that could possibly done to get the cost down. No one is saying that this is an easy problem.

Also you seem to forget that this is done, implemented, and priced. It is not just what SpaceX "thinks" what matters is what they did. Building a phased array of this size/type for < $1000 is another item on the list of things that people laughed at as impossible yet SpaceX has done it anyway. And following the usual pattern, you can now find articles accusing SpaceX of subsidizing these costs to the tune of thousands of dollars a piece. (I am sure initial units are subsidized in some way, since it is normal to use the volume production price while ramping up, but that is a different thing.)

More importantly "volume" for that device is on the order of 100s.

I don't think this is right - I think the existing volumes are in the millions.

Then you didn't read your own link. The in stock quantities are in the 100s. for the items that were non-zero. Buying large numbers of that part is not something anyone would do to make a phased array, unless maybe as a one-off.

[...] Starlink is making thousands of antennas with over a thousand antennas each. That is millions of antenna elements, and they are still at relatively low rate production. These are very different economies of scale and can easily change unit costs by orders of magnitude.

The generic learning curve is a 10% cost decrease for a 2x increase in volume.  So a million fold increase in volume (of the same part) would lead to a final cost of 35% of the original.   That's not nearly enough, which is why a custom design is needed, not just cheaper production of existing parts.

First of all, that rule of thumb isn't going to apply to simply anything. It certainly doesn't apply when you picked a bad starting point.
Also I have no idea how you did your math, but a million is about 2^20, and 0.9^20 is 0.12, Not that that number is relevant. When you double 20 times, you use such different techniques that no rule of thumb will help.

The other minor caveat is that I think the indications have been that they are building these themselves, the scale they want is certainly at the level to justify buying their own machines to dedicate to the fabrication.

This I strongly disagree with.  A modern semiconductor fab costs 10s of billions of dollars, and needs thousands of specialized workers.  Their only hope, cost-wise, is to piggyback on someone else's high volume process.  So SpaceX will do the design of the chip, but one of the established semiconductor fabs will do the manufacturing.

Where in the world are you getting these numbers? My college had some robust semiconductor fabrication, it certainly didn't have thousands of people operating it or pay billions of dollars. (Yes the equipment was probably 2nd hand in general, but it is not like SpaceX needs to use 7nm transistors or something, so they quite possibly could get what they need 2nd hand as well.) They shouldn't need a building full of machines for their scale, but keeping a small number constantly occupied is something that makes sense at their scale, and that is the scale where you are probably better off buying your own.

[Robotbeat]

They may be doing all the development themselves, but they shouldn’t have to actually fab it themselves. Like Nvidia, doubtless they’ll just get TSMC to make them.

If they’re doing a mostly-analogue chip, it could make sense to do the fab themselves as they won’t get much advantage from extremely small feature size that pushes you to go with a state of the art fab.

[ncb1397]

Also you seem to forget that this is done, implemented, and priced. It is not just what SpaceX "thinks" what matters is what they did. Building a phased array of this size/type for < $1000 is another item on the list of things that people laughed at as impossible yet SpaceX has done it anyway.

Umm, there is no real proof that they built these antennas for < $1000. There appears to only be a few thousand antennas in the wild, as such they could easily have subsidized the price as many companies do (printer companies make money on ink, cell phone service providers sell ~$600 phones at ~$200,  etc.). The fact that they aren't providing cheaper plans than the $100 per month like other satellite internet service providers do suggests to me that there is some subsidization.

They may be doing all the development themselves, but they shouldn’t have to actually fab it themselves. Like Nvidia, doubtless they’ll just get TSMC to make them.

If they’re doing a mostly-analogue chip, it could make sense to do the fab themselves as they won’t get much advantage from extremely small feature size that pushes you to go with a state of the art fab.

Think it isn't TSMC...

GaAs monolithic microwave IC maker WIN Semiconductors has become a supplier of satellite communication ICs used in SpaceX's Starlink, according to industry sources.

https://www.digitimes.com/news/a20200820PD207.html

[meberbs]

Also you seem to forget that this is done, implemented, and priced. It is not just what SpaceX "thinks" what matters is what they did. Building a phased array of this size/type for < $1000 is another item on the list of things that people laughed at as impossible yet SpaceX has done it anyway.

Umm, there is no real proof that they built these antennas for < $1000. There appears to only be a few thousand antennas in the wild, as such they could easily have subsidized the price as many companies do (printer companies make money on ink, cell phone service providers sell ~$600 phones at ~$200,  etc.). The fact that they aren't providing cheaper plans than the $100 per month like other satellite internet service providers do suggests to me that there is some subsidization.

They'd have to be idiots to subsidize to that extent, they have to pay off the cost of building out their constellation, and that is more than enough reason for the pricing. And you are continuing to ignore the fact that this service is comparable to cable in terms of both service level and price, yet is available to many people who don't have anything close to cable levels of service. It makes no business sense to price their service below the cost of cable, especially now as they are still building out the constellation. (It is another story if you talk about other markets outside the U.S. where either internet is cheaper, or people can't afford as much, by that point the marginal cost of adding subscribers in new geographic areas is low. They have already indicated that prices would vary based on local markets.)

Again, the only significant subsidization that makes business sense is the standard selling low rate initial units for the projected full rate pricing. I doubt they have profit built in to the terminals themselves, since the profit (and paying for the satellites and launches) would be built into the monthly cost.

Seriously, this is the same worthless pattern of FUD and denial that happened with SpaceX's original F9 prices, reuse, etc.

[DistantTemple]

Also you seem to forget that this is done, implemented, and priced. It is not just what SpaceX "thinks" what matters is what they did. Building a phased array of this size/type for < $1000 is another item on the list of things that people laughed at as impossible yet SpaceX has done it anyway.

Umm, there is no real proof that they built these antennas for < $1000. There appears to only be a few thousand antennas in the wild, as such they could easily have subsidized the price as many companies do (printer companies make money on ink, cell phone service providers sell ~$600 phones at ~$200,  etc.). The fact that they aren't providing cheaper plans than the $100 per month like other satellite internet service providers do suggests to me that there is some subsidization.

They'd have to be idiots to subsidize to that extent, they have to pay off the cost of building out their constellation, and that is more than enough reason for the pricing. And you are continuing to ignore the fact that this service is comparable to cable in terms of both service level and price, yet is available to many people who don't have anything close to cable levels of service. It makes no business sense to price their service below the cost of cable, especially now as they are still building out the constellation. (It is another story if you talk about other markets outside the U.S. where either internet is cheaper, or people can't afford as much, by that point the marginal cost of adding subscribers in new geographic areas is low. They have already indicated that prices would vary based on local markets.)

Again, the only significant subsidization that makes business sense is the standard selling low rate initial units for the projected full rate pricing. I doubt they have profit built in to the terminals themselves, since the profit (and paying for the satellites and launches) would be built into the monthly cost.

Seriously, this is the same worthless pattern of FUD and denial that happened with SpaceX's original F9 prices, reuse, etc.

This is not FUD and denial. Loads of major technological things are incredibly expensive at first, and the price drops as the market explodes. Hard Disk drives! CD's, CDRW and more recently Blue Ray. All were stupid prices at first. ANd remember the cost of RAM in 1998! So the "Idiots" line is nonsensical!
Its obvious that Starlink is onto a slam dunk win. Despite the government programs to connect rural populations to the internet, there are BILLIONS with poor or zero internet. No other provider is near offering what SpaceX is doing. And populations are beginning to hear the news, and longing for it to reach their country. Regulators and legislators around the world will have to fall into line and authorise it! People on here are talking in terms of 10's of millions of subscribers..... and are wrong. within a couple of years SL could be in the 100's of millions, if they can produce the antennas fast enough, and manage the backbone etc.
So if SX "blew" $1B on "subsidising" these early versions of the antenna, it would be money well spent, and easily recouped before 2024!   There have been endless discussions here of what the "cost" of an SL antenna is anyway. It is nonsense to pass on full "costs" of this first tranche to subscribers. Plus all these BtNBeta participants are doing SX's testing... and paying SX for the privilege!
Remember EM thinks in terms of "Giga Factories". 
Also we have seen he will purchase companies to get their expertise, IP, staff, and 100% of their production capacity!
That may be an option here to quickly scale production!
Still I don't work in this industry, so this is just my opinion.
Apart from the "Idiots" most of the rest of what you say meberbs makes sense, and is a more reasoned response to the previous post

[Robotbeat]

Reminds me of every discussion of SpaceX’s low launch costs. “They must be subsidizing it with Musk’s money!”
“NASA is subsidizing Air Force Launches!”
“Air Force is subsidizing commercial launches!”
“Commercial launches are subsidizing NASA launches!”
“It must be an investor scam!”

Or, you know, they figured out a way to cut costs by a lot.

I mean, I wouldn’t be surprised if $500 per unit doesn’t fully cover costs for these first few thousand units, but SpaceX isn’t dumb. They’re not going to lose massive amounts of money when they are heavily supply constrained and have demand coming out the wazoo. They could charge thousands of dollars for these things and still have demand.

[meberbs]

This is not FUD and denial. Loads of major technological things are incredibly expensive at first, and the price drops as the market explodes. Hard Disk drives! CD's, CDRW and more recently Blue Ray. All were stupid prices at first. ANd remember the cost of RAM in 1998! So the "Idiots" line is nonsensical!

What in the world are you talking about? You seem to be addressing something completely unrelated to what I wrote. Things that used to be sold for higher prices now being sold for lower, would only help the argument for SpaceX's long term goals, but it in no way justifies them throwing money out the window.

Its obvious that Starlink is onto a slam dunk win. Despite the government programs to connect rural populations to the internet, there are BILLIONS with poor or zero internet. No other provider is near offering what SpaceX is doing. And populations are beginning to hear the news, and longing for it to reach their country. Regulators and legislators around the world will have to fall into line and authorise it! People on here are talking in terms of 10's of millions of subscribers..... and are wrong. within a couple of years SL could be in the 100's of millions, if they can produce the antennas fast enough, and manage the backbone etc.

These are things you might need to remind ncb1397 of, but not me.

So if SX "blew" $1B on "subsidising" these early versions of the antenna, it would be money well spent, and easily recouped before 2024!   There have been endless discussions here of what the "cost" of an SL antenna is anyway. It is nonsense to pass on full "costs" of this first tranche to subscribers. Plus all these BtNBeta participants are doing SX's testing... and paying SX for the privilege!

No really, it simply does not add up. Throwing billions at subsidizing the user terminals does not make sense. You can't assume that users will end up being with the service for 5 years, some will move, some antennas will fail or be damaged, and meanwhile there is the cost of the satellites that they need to make their money back on. As Musk has said, the goal is to not go bankrupt. You don't achieve that by selling your product at what would be a loss even at full rate production when you haven't even begun paying off the large fixed costs of the satellites. The satellites are planned to live for 5 years, and they need to therefore pay for building and launching 4000-12000 satellites within 5 years. That will not work if it takes a year or more just to have users pay off the terminals.

And as I stated multiple times this is very different than the typical practice of not making the early (beta) users amortize all of the development and setup cost during low rate production, the price should represent roughly the expected cost during high rate production. I went out of my way to repeat this and you still manage to twist what I said completely.

Remember EM thinks in terms of "Giga Factories". 
Also we have seen he will purchase companies to get their expertise, IP, staff, and 100% of their production capacity!
That may be an option here to quickly scale production!
Still I don't work in this industry, so this is just my opinion.

Again, you are addressing the wrong person here. These are more things that explain how it really is reasonable that they have driven the costs of the terminal down far below what some "experts" expected. We know that they have started out at low rate production, but this discussion is about the cost at full rate production, which they are in the middle of ramping towards, and the value that is reasonable to expect them to assign to the terminals.

Apart from the "Idiots" most of the rest of what you say meberbs makes sense, and is a more reasoned response to the previous post

I don't think you read that part of my post right, it was ncb1397 who implied that SpaceX is using a terrible strategy of losing money on every subscriber they pick up. He has a history of similar terrible comments, misrepresenting SPaceX prices and costs, etc. This is literally the exact same claim that was falsely made about F9 pricing.

[freddo411]

I've have a starlink invite for my new place in MT.    There is a deadline to sign up (Nov 20th for me).    I won't be at my new place for 2 weeks after that.   

Question for the group.   Those of you that sign up and got a dish delivered.   How quickly was the dish delivered after you signed up?

Turns out you can have a "service address" where you use the dish -- AND have the dish actually shipped to another address.   My problem solved.