Literally Tearing Apart A SpaceX Starlink Antenna

While SpaceX’s constellation of Starlink satellites is nowhere near its projected final size, the company has enough of the birds zipping around in low Earth orbit to start a limited testing period they call the Better Than Nothing Beta. If you’re lucky enough to get selected, you have to cough up $500 for the hardware and another $100 a month for the service. Despite the fairly high bar for getting your hands on one, [Kenneth Keiter] decided to sacrifice his Starlink dish to the teardown Gods.

We say sacrifice because [Kenneth] had to literally destroy the dish to get a look inside. It doesn’t appear that you can realistically get into the exceptionally thin antenna array without pulling it all apart, thanks in part to preposterous amount of adhesive that holds the structural back plate onto the PCB. The sky-facing side of the phased array, the key element that allows the antenna to track the rapidly moving Starlink satellites as they pass overhead, is also laminated to a stack-up comprised of plastic hexagonal mesh layers, passive antenna elements, and the outer fiberglass skin. In short, there are definitely no user-serviceable parts inside.

The dish hides many secrets under its skin.

Beyond attempting to analyze the RF magic that’s happening inside the antenna, [Kenneth] also takes viewers through a tour of some of the more recognizable components of the PCB; picking out things like the Power over Ethernet magnetics, a GPS receiver, some flash storage, and the H-Bridge drivers used to control the pan and tilt motors in the base of the dish.

It also appears that the antenna is a self-contained computer of sorts, complete with ARM processor and RAM to run the software that aims the phased array. Speaking of which, it should come as no surprise to find that not only are the ICs that drive the dizzying array of antenna elements the most numerous components on the PCB, but that they appear to be some kind of custom silicon designed specifically for SpaceX.

In short, there’s still plenty we don’t know about how this high-tech receiver actually works. While [Kenneth] does a respectable job of trying to make sense of it all, and we admire the dedication required to rip apart such a rare and expensive piece of kit, it’s still going to be awhile before the hacker community truly masters the tech that SpaceX is putting into their ambitions global Internet service.

[Thanks to Alex for the tip.]

85 thoughts on “Literally Tearing Apart A SpaceX Starlink Antenna

    1. Hey, thanks for making this video. It was REALLY informative and extremely well done. I am hoping to learn more about the the actually phased array antennas. I have a suspicion that this is a product that was developed by a company in Ottawa Canada call C-COM Satellite ( ).

      They have spent several years funding and working with the University Of Waterloo ( in Ontario, Canada) who has a global center of exellence in RF antenna development. If you read their public disclosure, they talk about the ongoing testing of their plannar phased array antenna for use with low earth orbit satellites in the Ka microwave spectrum.

      I am not sure if they have licensed this technology to (I believe you said ST Micro was the silicon provider) but if anyone has any ability to validate that this might be the C-Com Satellite IP I would GREATLY appreciate a response.

      Full disclose, I used to be a professional hedge fund manager and I have invested a significant amount of my own money into C-COM’s stock which trades on the Toronto Stock Exchange under the symbol CMI

      If this is what I think it is, once it becomes widely known I expect either Starlink to buy C-Com or someone else will. The plannar phased array antenna are going to transform global high speed IP internet access and C-Com will be worth billions.

      If this is what it is, this is my early retirement secured.

      ANY HELP would be greatly appreciated. You can also email me directly at

      Thank you all in advance and for this amazing video.

    1. Much better than I expected. Look at all those elements! It must be one heck of a beamformer. This is why they say it is better than military jet fighter RADAR. Hmmm. I can see it now. A future HaD blog entry “Multi-Target Long Range Tracking RADAR from a Repurposed Starlink Antenna and an SDR Dongle for your Phone.”

      Also good for home built airplanes or pointing back at your own ground station or drone from a moving vehicle. I see an explosion of moor cool applications just around the corner.

      Give the kids cheap SDR (And Matlab, Jupyter Notebook, and all that.) and 6 years later this is what you get!

      1. Yeah, it might be a great base for amateur phased array experiments. Saves a lot of design complexity and cost. Hacking might be hampered by the fact there are no freely usable/ISM bands around 10 GHz. Not sure about radars, it is probably too noisy and and low power – range is proportional to power^4. FMCW techniques are an option if the frontend is suitable. Cooperative target tracking too, if the target has some sort of transponder.

        Did they ever clarify what they meant by saying it’s better than fighter radar? I have a hard time believing that. Maybe when considering relatively small antenna size and cost, but not in terms of state of the art performance. Airborne ~10GHz radars are larger and have a bit more elements so naturally have a tighter beam and better sidelobe performance. AESA arrays typically have one TR module per element, that does wonders for sidelobes and multi-beam/multi-frequency performance too.

        1. there is no way in hell it’s better then a 5000-ish element active array radar like the AN/APG-81 that alone is worth well over a megabuck…also this particular radar probably beats it in the data transfer speed as well if used in that mode :D
          (point-to-point mode only though)

          1. I took them to mean for its size and number of elements. Like the way car manufacturers always find some metric that makes their car “better” than the competition.

  1. Each block has a “TP_VFEM_XX” testpoint near it, “FEM” probably indicates “front-end module”, which could be as simple as a matched LNA and switch, or could be an entire downconvert chain. The “LO” in “LO_CLK” is most likely “local oscillator”, not “low-frequency.

    1. It almost certainly has the primary mixers + LO multipliers inside. This really really simplifies board design, as the frequency is way lower. In addition, you need only to make one ic with an EHF process, and you order many since there is one per element. It saves cost.

      For such a large array you don’t need very high quality RF components either. Some implementations use only 1 bit digital per fem.

      1. How do they do it at this frequency and what freq would the LO be? I have assumed that just like SDR front ends, they switch or chop the input at the LO rate, which is the same as multiplying by all the odd harmonics of the switching rate. But what is the chop freq? A cheap HC5053 multiplexer chip can go 200MHz. Do you need an LO higher than that?

  2. I’m working on it – it’s a filler word, and I know it bugs people. Generally, I’ve noticed that I use it when I’m searching for the next word, as it provides perceived continuity that prevents a short pause from being misinterpreted as the end of a sentence. You can certainly tell when I’m reading from the teleprompter 😁

    Luckily, I don’t think I’m in danger of becoming a professor anytime soon – but thank you for your feedback! 👍

    1. That is good. Equip a few audience mebers with little clickers, with a WiFi capability, you wear a little tingler-shocker, amazing how you will get up to speed – in jest…. We had one prof who went into a rage when we counted his incessant ummm phrases, He even complained to the admins and other profs, who said ‘wear it, or stop doing it’. It also aggravated almost all other staff members from students to post docs.
      He had a sense of grandeur, he was a prof, you = the great unwashed. He did not realize that in Engineering Physics at U of T he was facing many brilliant minds – and fools were not born gladly.
      You are correct – it is an interposing word that lets you riffle for the right one. Lecture planning helps, but you will have to improve your cross linking on-the-fly. If the problem is intractable, script the entire talk, and condense it as well – you sould be able nip and tuck this into a 10-12 minute exposition and still mention all your material.
      You might search utube or google for hints and kinks.

      1. We had Prof. that stared at the ceiling the ENTIRE time he lectured and never looked at his students. That ended the day a Playboy centerfold was stuck on the exact spot he would look at. He never did it again and said nothing about it.

    2. Ken, good on you for taking William’s comment as well as you did. I would have either ignored him, or given a nasty response at the risk of having my comment deleted, because despite saying it’s not an attack, to me he came across as a jerk.

      1. Agree. It was very very jerky. Most of us are EXTREMELY aware of our ticks. Pointing them out and offering stupid advice doesn’t help in any way. William, maybe you should take this as a wake up call. Speaking to others like that will bring you more pain than pleasure.

  3. Read the patents first. Spacex generally doesn’t patent things on the theory that the Chinese will just steal it anyway. But, there are a good number of patents granted on that antenna and probably more pending. They’re probably counting on the proprietary aspects of the technology and manufacturing process to deter knockoffs. That thing runs in the 40-50GHz area. Far higher than most satellites and download speeds are in the 120 megabit area. Just tearing it apart will tell you less than reading the patents.

    1. Economics of scale, among other things. A generation or two down the road you might reclaim these early antennas and reuse them in impoverished areas and bring down the price of newly manufactured units to reasonable levels for nonprofit or subsidized distribution.

    2. What? You thought the old stick with a rock on a string would work? And they need one per village, not one per computer. Or one per remote farm or mine or island or long night north of Yellowknife.

    3. First off, it’s a “better than nothing” beta, as mentioned in the post, so it’s still in the VERY early stages. It’s safe to assume the prices will come down with time. And of course it’s a proprietary dish; there’s no currently suitable device on the market.

      1. The ‘Me First’ people would pay any price, since they are ‘better’ than most of us… No offense to the author, for making the sacrifice. The beta units are going to be extra rugged, to discourage the curious, but also it needed to be. This same antenna design, will be used to test any, and all possible applications of Starlink. They are going to want them mounted on vehicles, and bounced around out in the wild. They need to survive a BLM ‘peaceful’ protest. What they really want tested, is the internet, anywhere, anytime aspects. There will likely be several consumer versions, when they go to production, based on the intended use, and different prices. It’s usually the ‘service’ that generates the profits, the hardware gets paid for in the package.
        120 megabits is pretty fast, for a single user, who would be thrill, just to get some sort of internet. One access point, could feed dozens of users, with reasonable speed, if stream HD content isn’t their main goal.

        1. “One access point, could feed dozens of users, with reasonable speed, if stream HD content isn’t their main goal.”

          Still have my POTS office modem from back in the day. This will kind of be a modern version.

          1. The load shouldn’t overwhelm the capacity that easily as no user is using more than a tiny fraction of the total satellite to reach their end goal (probably 2 maybe 3 – and possibly only 1 to a ‘nearby’ ground station with its massive conventional pipes). But if an area is saturating the satellites that will be in their visible sky it will be ultimately add more/replace with better satellites. Same as any system, saturated fiber link between nations add in another etc..

      2. You can bet that the $500 nominal price is not what they cost to make. I once had a FWA phone installed at home, current dev cost was £30M and rising, and I was the 20th subscriber, so it cost over one meeeeleon pounds!!!

        1. I was one of the 1,000 Sprint Ion customers. Sprint literally spent a billion dollars to acquire the 1,000 of us, so my share was $1 million. It cost a lot of money to build high-tech infrastructure, and I’m sure Starlink is pretty much the same.

    4. And you expected what? An ethernet cable dangling from the satelite? Using your DirectTV dish and moving it really quick to aim at the moving satelites? It was clearly stated from the very beginning you will need their dish.

      1. This is likely exactly the plan. For a small village who didn’t have any internet at all. 100mbps shared among 100 people would probably be a godsend compared to what they have now if anything. You could install something like this at the village school, prioritize the traffic for the school and offer what remains to the public in the village. Likely during the day when students might be making heavy use of it, the adults would be doing other work that doesn’t require internet, then in the evenings when school is out all the bandwidth becomes available to the general public for leisure time use of the internet. Depending on the density of the village a starlink dish and 4-5 mesh routers would likely provide amazing coverage.

    5. Even if you assume that each array costs a few thousand which is well over the top its still potentially millions cheaper than trying to run similar speed broadband. You can drop heaps of relatively cheap receiver dish in every remote location and just wire up the local area – which is much cheaper per mile covered than the high bandwidth backbones for a real wired infrastructure – and you don’t need to run 100’s of miles of it to get between the villages/towns..

      Also have to point out that to huge swathes of the world the bandwidth of one of those dishes is greater than they really need – until you spend all day streaming at 4K you really don’t need much, just accessing text based websites even ye ol world dialup is use able and probably massively quicker than any connection they can get (as having to drive/ride/walk hours to get to the nearest uplink really slows the data rate, even if when they finally get access its 20x faster…)

    6. How much did a C64 cost when it first went on sales? Here is was $1,299 CDN. How much did it cost when it was discontinued? $99.
      How much for an small micro with 64KB of ram today? Less than $10.

    7. Um: as usual, there are “reasons and REAL reasons”… reason: “we do it for the advancement of mankind”. REAL reason: “let’s make a decent bit of money – but over a loooong period of time” = tons and tons of cash. The reasons-and-real-reasons principle can be seen everywhere in politics and big business and the Orwellian double-speak that has crept into our language and our minds only supports this way of doing business. Take children for example (I used to work for a biiiig toy company): “children are our future!” … yes, but first and foremost they are consumers and “influencers” (McNeal – The Kids Market), which has a huge impact on politics and media content (often with severe negative long-time consequences for them).

      Now take poor people or underdeveloped regions: a great opportunity for politicians to play the “justice” card and a huuuuge opportunity for companies to build propositions around what’s called the “sachet marketing” principle: selling Western shampoo brands at 50 US cents for a single-dose pack to a poor woman in India 6x per year is a very lucrative business model. Big brands (FMCG) love poor people – simply because there are so many of them and their behavior is very predictable. As most spending these days is on media / media services, “antennas for the poor” are mainly one thing: a great way to make a lot of money.

      1. I think Elon saw this as an area of possible incursion (disruption) by the Starlink concept. The first-cost of ~~$2200 per terminal was to be intentionally borne as the cost of this disruption. I expect Starlink engineers are hard at work making a future series of terminals at a far lower cost. I have yet to see a tear-down in detail that would allow for a detailed part-by-part cost analysis. While making a lot of money is a valid modus operandi, so is widespread satellite based bidirectional internet communications also valid. I anticipate the large volume of potential terminals to be hopefully built and sold will fuel this terminal cost decline – rate of decline??
        As it sits the $500 plus $100/month is bearable in many areas that are now badly serves, this)assumed) cost reduction will allow spread to many very poor areas where even $500 +$100/month is too much. Many of thoise areas have economics based zero HS internet – 56K modems being their max and many not even that.
        So this is a book that remains to be written – one I hope to read about soon.

    8. Remote poverty-stricken areas are not completely bereft of resources (if only indirectly). India, for example, has been trying to connect its remote areas with fiber for 10 years now, at a budget of more than $5 billion. Of course ground infra is hard, especially when it’s a government agency that’s running point – but spend a fraction of this on one or a couple of dishes per village and you’ll have working internet yesterday.

      Even at full price, $500 is only $1/month if amortized across 10 households and 4 years.

    1. Elon is a good business man. His accomplishments are built on the backs of engineers that give the best years of their lives, if not their entire lives, to the company. I hope that the people in those positions can look back on their choices and be happy. I know I couldn’t; that is why I left the ‘high end’ of the tech industry.

      It was the same thing with Apple for years. Jobs this and Jobs that. He might have had a vision, but it was the overworked and under appreciated engineers delivering.

  4. the technology in here also represents, so some degree, the technology of their future Starlink Mesh technology which will certainly be state of the art stuff that their bottom line depends on.
    don’t want to be giving that away to other wanna-bees.

  5. It looks like the individual elements are electromagnetic band gap antennas the circular patch excites the notched patch. The honeycomb ensures the gap between the excited patch and the circular patch. My guess for why the notched patch is not coaxially driven could be impedance or bandwidth. The 8 antenna clusters would appear to be a steerable beam channel which in turn is steered to lock onto one satellite. My bet is this thing can lock onto as many as 64 satellites. Those special ST chips are beam steering + frontend + demodulation. The actual rf chain is kept minimal by this arrangement. The processors for each channel would send the decoded data on to the main processor appears to assemble it into a TCP/IP packet stream. The special sauce here is likely the coordination of user stations to use certain satellites and the switching between satellites as the wiz by. The thought of all of load leveling and routing of data is a bit boggling. My bet is the user stations receive commands from the network that is aware of all of the user stations that are online and where they are located.

  6. The antenna is a dual polarization dual feed patch microstrip antenna, the chips are a transceiver chips for signals, each chip supports 4 or 8 polarization ports (1/2 * antenna number), the chips are from
    its very simple RF system, but the beam forming and tracking of satellites overhead passing by is what is complicated and done in the modem.

    The antenna details are shown in their patents, at

  7. Coming along 18 months later, Starlink’s gaining a foothold among ham friends living out in the wilds, I found and marvel at Kenneth’s teardown. It’s clear to me. It’s just a scaling up of airliner TV/Internet CP patch array tech, pioneered by Conextant in the ’90s, then Rockwell, Boeing, whoever. Nothing really new, but nicely done. Yeah, I blurb. Who says consumer isn’t state of art? Consider your phone with dirt cheap 4-8 channel MIMO 5 GHz antennas in your hand for $400, whole thing. EHF makes it even easier to scale up, that aluminum shield probably just to protect a user under it from maybe 20W max ERP. Power-controlled links of course. Planar is actually easier than conformal. I see 2×4 element groups for TX and RX side by side, RHCP and LHCP resp. Then ~ 40 groups, TX and RX each, over the whole array. Yep, nice multilayer board. CP patches are broadband; maybe even one size fits TX and RX, GHz apart! It’s just element pattern x group pattern x array pattern; simple trig, see any text. I see a two element Yagi-Uda, driven and director, at each patch site for a bit more gain; clever, but possibly unpatentable as this is as old as Shintaro Uda, 1926. Pound down Faraday shielding betwixt everything. Viola! But what a beautiful, straight-talking video. Super congrats Kenneth!

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.