Starlink Internet Connection equipment - Home/Office user

[groknull]

Some speculation and observations about potential structures of the Starlink user terminal phased array antenna assembly.  These speculations are based on the features seen in the user terminal teardown video posted upthread.
https://forum.nasaspaceflight.com/index.php?topic=50047.msg2158668#msg2158668

Attached is a notional model of a possible structure of a phased array antenna cell that, to a first approximation, seems to match features seen in the teardown video.

This model is intended to be a stepping off point for discussion.  All observations and criticisms welcome.

Observations:
- The antenna cell appears to be a stacked microstrip antenna.
- The patches on the second board have slots (known as Defected Microstrip Structures, IIRC) that may be for managing cross polarization characteristics as well as bandwidth.
- The Starlink antenna structure may be as simple as a 3 level stack - ground plane and driven element on the main board, and a parasitic element on the second board.  If that is the case, then the solid circular features on the board are the driven elements.
- There are stacked microstrip antenna designs with many more layers, some of which have multiple driven elements for multi-frequency operation.
- Defected Microstrip Structures are common in microstrip antenna designs.  Defected Ground Structures and driven elements with complex edge features are often used to tailor bandwidth, sidelobe generation, crosstalk, and other electromagnetic radiation characteristics.

Caveats, details and simplifications of the attached model:
- A stacked 4 element design with a single driven element is modeled.  Ground plane, driven element, solid parasitic element (all part of the main board), and slotted parasitic element on the second board.
- The ground plane is modeled as a simple planar surface with no defected structures except for the pass through for the driven element feed.
- The driven element is modeled as a simple square with featureless edges.  Many square driven elements actually have clipped corners or other cuts.
- The driven element is modeled with a centrally located feed to simplify the model.  Many patch feeds are off center.  Some driven elements have two or more feeds to create alternate or circular polarizations.
- The driven element is shown twice, the one two the left is flipped to show the feed on the bottom.
- PCB layers A, B & C are the top three layers when viewed from the antenna (not chip) side of the board.  Layer C is the visible one.
- Colors are arbitrary.
- Dimensions are approximations based on eyeballing images from the video.  PCB layer and trace thicknesses were picked to be recognizable, and are probably not proportional to other features.

Edit: typos

[LouScheffer]

Upon closer inspection, I'm changing my estimates.  By counting, there are only 1563 antennas on the board.  Therefore each small chip cannot drive 3 antennas.   The same results can be obtained by superimposing the component side and antenna side, which shows each small chip drives 2 antennas.  It also shows that around the edges, many elements have no corresponding small chip.  This makes sense - you would want the outer edges antennas to be dummies, so that every active element is surrounded by 6 other antennas and sees the same environment.   Also, the digital circuits take some room, and it looks like the RF spots under them are also not driven.

So overall, I now think there are 72 big chips, each driving 8 small chips, each driving 2 antennas, for a total of 1152 active elements in the array.  The others, all round the edge, are dummies to preserve a uniform electromagnetic environment for the active elements, and to make room for the other needed circuits.

I counted 1336 antennas.  (Double checking of counts and math welcome.)
Bottom group (344):
- rows of 8 to 18 antennas (incremented by 1)
- rows of 17 to 21 antennas (by 1)
- rows of 20 to 22 antennas (by 1)
- rows of 21 and 22 antennas

Center group (648):
- one row of 23 antennas
- 15 pairs of rows with 22 and 23 antennas in each row pair

Top group (344):
- same as the bottom group

The video indicates 79 larger chips, not 72.  I counted 79 as well.

That allows 632 smaller chips.  If two antennas per smaller chip, then 1264, which agrees well with 1336 minus border cells.  As you (I think) pointed out, some larger chips may have fewer than 8 smaller chips attached to them.

I agree that some of the edge cells are likely to be passive/grounded or parasitic to modify the beam edge characteristics.

Upon recount, I get 1463 antennas.  I've attached the picture so others can check.   From the numbers in each region (defined by blue lines), I get 8*100+90+14+66+77+15+50+24+45+36+78+57+10+86+15 = 1463 antennas.

The antennas covered with black lines are those that do not have a full complement of 6 neighbors.  There are 132 of these, which I'm sure will not be used, leaving a max of 1463-132 = 1331 active antennas.  Others may be unused as well for real estate reasons, or if one row of unused cells around the rim is not enough electrically.

[LouScheffer]

The video indicates 79 larger chips, not 72.  I counted 79 as well.

That allows 632 smaller chips.  If two antennas per smaller chip, then 1264, which agrees well with 1336 minus border cells.  [...]

I agree that some of the edge cells are likely to be passive/grounded or parasitic to modify the beam edge characteristics.

OK, I agree about the 79 larger chips, see photo below.

So overall, 79 x 8 x 2 = 1264 active elements.

[groknull]

Upon closer inspection, I'm changing my estimates.  By counting, there are only 1563 antennas on the board.  Therefore each small chip cannot drive 3 antennas.   The same results can be obtained by superimposing the component side and antenna side, which shows each small chip drives 2 antennas.  It also shows that around the edges, many elements have no corresponding small chip.  This makes sense - you would want the outer edges antennas to be dummies, so that every active element is surrounded by 6 other antennas and sees the same environment.   Also, the digital circuits take some room, and it looks like the RF spots under them are also not driven.

So overall, I now think there are 72 big chips, each driving 8 small chips, each driving 2 antennas, for a total of 1152 active elements in the array.  The others, all round the edge, are dummies to preserve a uniform electromagnetic environment for the active elements, and to make room for the other needed circuits.

I counted 1336 antennas.  (Double checking of counts and math welcome.)
Bottom group (344):
- rows of 8 to 18 antennas (incremented by 1)
- rows of 17 to 21 antennas (by 1)
- rows of 20 to 22 antennas (by 1)
- rows of 21 and 22 antennas

Center group (648):
- one row of 23 antennas
- 15 pairs of rows with 22 and 23 antennas in each row pair

Top group (344):
- same as the bottom group

The video indicates 79 larger chips, not 72.  I counted 79 as well.

That allows 632 smaller chips.  If two antennas per smaller chip, then 1264, which agrees well with 1336 minus border cells.  As you (I think) pointed out, some larger chips may have fewer than 8 smaller chips attached to them.

I agree that some of the edge cells are likely to be passive/grounded or parasitic to modify the beam edge characteristics.

Upon recount, I get 1463 antennas.  I've attached the picture so others can check.   From the numbers in each region (defined by blue lines), I get 8*100+90+14+66+77+15+50+24+45+36+78+57+10+86+15 = 1463 antennas.

The antennas covered with black lines are those that do not have a full complement of 6 neighbors.  There are 132 of these, which I'm sure will not be used, leaving a max of 1463-132 = 1331 active antennas.  Others may be unused as well for real estate reasons, or if one row of unused cells around the rim is not enough electrically.

The numbers on your annotated antenna count image need to be tweaked slightly:
66 in the upper left corner should be 67 (+1).
86 in the lower left corner should be 84 (-2): the element at the extreme corner is not an element - easier to see a few frames later, and that section is hard to count (at least for me).
78 in the lower right corner should be 77 (-1).
10 in the lower left corner should be 11 (+1).  I kept missing that last element because the eye is drawn to the line of 10.
57 in the lower right corner should be 59 (+2).

This nets out to +1, or 1464 elements.

This matches to the per row incremental element recount I did.  I clearly miscounted, and made embarrassing basic arithmetic errors during my previous effort.

The element array is mirror symmetric (numerically) on both axes.

For those who want to handle this programmatically, here is an array of the number of elements in each row, grouped by discontinuities:
[8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18,
17, 18, 19, 20,
19, 20, 21,
20, 21, 22,
21, 22,
23, 22, 23, 22, 23, 22, 23, 22, 23, 22, 23, 22, 23, 22, 23, 22, 23, 22, 23, 22, 23, 22, 23, 22, 23, 22, 23, 22, 23, 22, 23,
22, 21,
22, 21, 20,
21, 20, 19,
20, 19, 18, 17,
18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8]

Note that these numbers are based on the orientation of the antenna side images from the teardown video.  The chip side is rotated 90 degrees.

[LouScheffer]

The numbers on your annotated antenna count image need to be tweaked slightly:
66 in the upper left corner should be 67 (+1).
86 in the lower left corner should be 84 (-2): the element at the extreme corner is not an element - easier to see a few frames later, and that section is hard to count (at least for me).
78 in the lower right corner should be 77 (-1).
10 in the lower left corner should be 11 (+1).  I kept missing that last element because the eye is drawn to the line of 10.
57 in the lower right corner should be 59 (+2).

This nets out to +1, or 1464 elements.

This matches to the per row incremental element recount I did. 

Agree, with the minor proviso that I have 2 regions of 10+1, where you counted one region of 11.  So count is preserved.   Here is the corrected image.

[LouScheffer]

Also, the teardown guy was kind enough to hold a measuring tape across the array.

From his shot with the tape horizontal, we see that 21 unit cells (each cell measured from the 4:30 to 7:30 position, in clock coordinates) span from 4.3cm to 48.8 cm (ignoring the cell under his thumb which is hard to measure).  That gives a cell size of (48.8-4.3)/21 = 2.119 cm, or 21.19mm.   We can convert this to a center-to-center spacing of 21.19/2/cos(30^o) to get a center-center spacing of 12.23 mm.   

The diameter is less accurately measured since we can't sum across unit cells, but looks to be about 6 mm.

The spacing makes the cells lambda/2 apart at 12.25 GHz.  This is about what you would expect, as the array should be sized for the higher transmit frequency.

EDIT:   It looks like downlink is 10.7-12.7 GHz, and uplink 14-14.5 GHz.  So spacing is 0.436 lambda at 10.7 GHz, and 0.591 lambda at 14.5 GHz. 

[Robotbeat]

What's the lowest frequency these things can use?

[LouScheffer]

What's the lowest frequency these things can use?

The spacing does not determine the lowest frequency useable for an array.  Instead it's set by the lowest efficiency of the individual elements, and increase in beam width (as the face is fewer wavelengths across).  Since the lowest frequency used is 10.7 GHz, I'll bet it works well down to that frequency, and not much lower.

[LouScheffer]

We know SpaceX uses circular polarization from their FCC filings.

I can't see to top notched circle converting linear to circular polarization, because of its symmetry.  If it did convert one way, by symmetry it would also convert the other way, and the two would sum back to linear.  So I suspect the top notched circle is just to increase bandwidth, and maybe forward gain.  So how can the bottom circle generate circularly polarized signals?  It can do so with asymmetrical feed from the layers underneath, which we cannot see.  For example, the paper "Broadband Circularly Polarized Patch Antenna Arrays With Multiple-Layers Structure" does this by using an L-shaped feed on the layer below, and covers almost the exact frequency range needed (roughly 10-14 GHz), and is very close to the observed size.  I suspect this or some similar scheme is used.  Two figures from the paper are included.

[LouScheffer]

Some more back of the envelope calculations:

We know from FCC filings that the transmit power is 38.2 dbW, so 10^3.82 W, or 6600 W EIRP.  To get the EIRP per element, we need to divide by the number of elements (79*8*2) squared, so an EIRP of 4 mw per element.  Assuming they are radiating omni-directionally into half-space (probably they are a bit more directional) that's an RF power of 2 mw/element.   Assuming a drive impedance of 70 ohms (though this is far from a dipole, so I don't know how accurate this will be), then the peak-to-peak voltage is needed is sqrt(8*power*ohms) = 1.058 volts.  This is just about what you would expect from a CMOS antenna driver.

[LouScheffer]

How close can the beam point to the horizon?  Assume theta is the angle of beam from the surface, which is equal to the elevation if the surface of the dish is horizontal.  Assume we measure everything in units of lamba, the free space wavelength.

Now assume we are working in one of the 6 cardinal directions of the pattern.  From this direction, we see row after row of antennas, separated by 10.6 mm from the measurements from the teardown video.

The highest frequency used is 14.5 GHz, which has a wavelength of c/f = 20.69 mm.  So the elements spacing is 0.5123 lambda at this frequency.  Call this the spacing s.

We get constructive interference if the signal from the different rows is exactly 0, +- 1, +-2, etc. wavelengths.  So if we apply a delay D per row, we need s*cos(theta) -d = 0, +-1, +-2, etc.  If our desired beam is at angle delta, then we set the delay D=s*cos(delta).   Now we have s(cos(theta)-cos(delta)) = 0, +-1, +-2, etc.

Since |cos(theta) - cos(delta)| <= 2, if s < 1/2 then there is only one solution;  theta==delta and the only solution is the desired beam.  But here we have s = 0.5123 so in some cases multiple solutions are possible.  We'll take the case where the delay sums to -1 wavelengths (others are similar by symmetry).

Divide through by s to get cos(theta) - cos(delta) = -1.9512.   Since cos(theta) can get as small as -1, this will have solutions provided cos(delta) >= 0.9512.  This will happen if delta (the desired direction) is less than 18 degrees from the plane of the disk. 

So if the antenna tries to generate a beam at less than 18^o to the dish at the upper band edge, it will also generate a beam squirting off in the exact opposite direction at some low elevation.  Therefore, assuming a level dish, the minimum elevation angle that can be used is 18^o.

[gongora]

https://twitter.com/thesheetztweetz/status/1334495224899448839

The U.S. Patent Office this morning published 8 SpaceX filings for different parts and features of the Starlink antenna:

The links in the tweets didn't work for me, this should be one way to access them: patent office link

[su27k]

Short article based on the video: Starlink Satellite Dish X-Rayed To Unlock Rf Magic Inside

[dondar]

It appears to be a useless report because, for instance, it assumes a 17-23 Gbps per satellite bandwidth in 2028, even with the added v-band.  This seems unrealistic, given the SpaceX's demonstrated pace of iteration.

You're assuming the V-band constellation will actually happen.  I'm not so sure about that.  Regardless of the V-band situation, that report, which was commissioned by SpaceX competitors, is not so great.  It doesn't use the correct orbital planes (for either the existing or proposed constellation layouts).

it will happen "when V-transmitters are ready". Of course the FCC application deadlines will be reset. Just like I claimed in the beginning.
According to what I hear from ex-colleagues who landed in space comm industry the bottleneck is production capacity of antenna modules which is "bad". Very "bad". Hence everything is too expensive for SpaceX volumes. When this will be solved, you will see new job offers on SpaceX site.

[dondar]

Detroit area -

Just received invitation a few minutes ago and signed up immediately ($99 deposit). Received confirmation email - "Starlink will begin offering service in your area beginning mid to late 2021. Orders will be fulfilled on a first-come, first-served basis. You will be notified via email prior to shipment, and you will be charged the remainder of your balance once your kit ships."

I'm happy but very concerned about reports that the dishy requires 100W.  If that's continuous 24/7/365 I'll be signing off quickly.  Does anyone have any deeper info on power consumption?

That 100W sounds not much more than what my Dish/Router/Modem pulls right now for standard satellite, so it's competitive. I also think, but do not know for certain, that Starlink does not pull it's full wattage rating unless it is running its anti-ice heater.

But all in all, we're still talking to things in orbit here, so it's going to use some power.

My Starlink has averaged 98.4 watts over the last 11 days since I put it on the power monitor, with a peak draw of 185 W. It's been very cold, between -10 and 24 F, and snowy so I think the dish heater has been running most of that time.

it will be the same in the summer. (probably even higher consumption "because of physics"). Transmission modules generate plenty of heat. In fact Starlink dish is quite economical and is "top end" in the current design line.

[rsdavis9]

it will happen "when V-transmitters are ready". Of course the FCC application deadlines will be reset. Just like I claimed in the beginning.
According to what I hear from ex-colleagues who landed in space comm industry the bottleneck is production capacity of antenna modules which is "bad". Very "bad". Hence everything is too expensive for SpaceX volumes. When this will be solved, you will see new job offers on SpaceX site.

Is that spacex's production or suppliers production?
IE does spacex make their own antenna modules?

[oldAtlas_Eguy]

it will happen "when V-transmitters are ready". Of course the FCC application deadlines will be reset. Just like I claimed in the beginning.
According to what I hear from ex-colleagues who landed in space comm industry the bottleneck is production capacity of antenna modules which is "bad". Very "bad". Hence everything is too expensive for SpaceX volumes. When this will be solved, you will see new job offers on SpaceX site.

Is that spacex's production or suppliers production?
IE does spacex make their own antenna modules?

As far as I understand. The purchase the "pizza dish" mechanical parts "case" and mounting. As well as the cabling and the finished router/wifi box. The internals, the actual phased array and modem circuitry is supposedly manufactured by SpaceX. Plus the custom SpaceX designed chips are a special order which may be on a back order from the chip foundry right now just due to the basic backlog of the chip foundries.

[Asteroza]

it will happen "when V-transmitters are ready". Of course the FCC application deadlines will be reset. Just like I claimed in the beginning.
According to what I hear from ex-colleagues who landed in space comm industry the bottleneck is production capacity of antenna modules which is "bad". Very "bad". Hence everything is too expensive for SpaceX volumes. When this will be solved, you will see new job offers on SpaceX site.

Is that spacex's production or suppliers production?
IE does spacex make their own antenna modules?

As far as I understand. The purchase the "pizza dish" mechanical parts "case" and mounting. As well as the cabling and the finished router/wifi box. The internals, the actual phased array and modem circuitry is supposedly manufactured by SpaceX. Plus the custom SpaceX designed chips are a special order which may be on a back order from the chip foundry right now just due to the basic backlog of the chip foundries.

Yeah, the whole low end fab microcontroller shortage is having knock-on effects everywhere. SpaceX I think was contracting with STmicro for the antenna chips from Germany, but the question is if that order was sufficiently placed early enough to beat the all the rush ordering around the beginning of the pandemic and the Trump US semiconductor bans targeting chinese industries (this caused all the chinese smartphone makers to buy in bulk smartphone related chips to build buffer stock before the bans kick in, which caused all the fabs to prioritize those orders since they were being paid rush fees). The wifi router block is allegedly made by a Taiwan ODM , so their own chip shortages affect that too. I think it was leaked that Tesla is stopping one of their production lines due to shortage of components (modern vehicles use a ton of MCU's, and EV's even more so), in line with a number of other automotive manufacturers facing supply chain shortages.

The MCU shortage may actually be a more than expected existential risk for Starlink, as it could get strangled in the crib due to high user terminal cost (which currently is being heavily subsidized by SpaceX to increase the subscriber base), when all the stars (cheap launch costs, cheap sats, frequency priority) had otherwise aligned in their favor.

[RedLineTrain]

As far as I understand. The purchase the "pizza dish" mechanical parts "case" and mounting.

Judging from the job listings over the past year or so, SpaceX is surprisingly vertically integrated on this.  For instance, they were hiring for injection molding in Hawthorne.  I assume that they ultimately decided to buy rather than build on this type of stuff, but who knows.  They do have the old Triumph factory to fill out.

[oldAtlas_Eguy]

As far as I understand. The purchase the "pizza dish" mechanical parts "case" and mounting.

Judging from the job listings over the past year or so, SpaceX is surprisingly vertically integrated on this.  For instance, they were hiring for injection molding in Hawthorne.  I assume that they ultimately decided to buy rather than build on this type of stuff, but who knows.  They do have the old Triumph factory to fill out.

Even if most of the parts/components are manufactured by other companies. SpaceX/Starlink would still need somewhere to put the UT kits together. As well as likely the production lines for the guts of the phased array and modem that is in the antenna is likely to be manufactured in house need a place. With an eventual quantities of 1,000,000 units a year or 20,000 units a week. It will take a lot of floor space for booth the lines and the warehousing for the incoming parts and the storage space of the Kits prior to them being shipped out. The facility you mentioned would be great for that.