Daniel's on an Emirates flight somewhere over the Atlantic, and instead of watching a movie or taking a nap, he's running speed tests. He pulls a hundred and fifty megabits per second down with minimal ping and jitter, on his Android phone's built-in Wi-Fi. And then he sends us a whole list of questions about how that's even possible. He wants to know how in-flight Wi-Fi is actually set up, whether the antennas need special calibration for traveling at zero point eight Mach, what the wiring and access point placement looks like inside the cabin, why cellular routers for moving objects are often the same ones people use at home, and whether the gateway is where the real engineering constraints live. He also asks if they just stuck an AP every few dozen rows and hoped for the best.
Which is a completely reasonable question, because that's what it feels like when it works. You open your phone, you see the Wi-Fi icon, and you just... use it. But the fact that Daniel measured a hundred and fifty megabits with low jitter over the Atlantic tells you this is not a few APs thrown into an overhead bin. There's a real network design problem here, and it breaks into two layers. The air-to-ground link — or in this case, air-to-space — and then the cabin network that distributes that signal to three hundred people in a metal tube at thirty-five thousand feet.
And the thing Daniel's really poking at is which of those layers is actually hard. Because from the passenger seat, the cabin Wi-Fi feels like the obvious answer — you're in a long cylinder full of seats and bodies and luggage, of course the RF environment is a nightmare. But then you think about the antenna on the roof tracking satellites at combined closing speeds that would make a fighter pilot queasy, and suddenly the AP placement seems like the easy part.
Let's start with the antenna, because that's where the physics gets genuinely wild. Daniel asked whether the antennas need to be calibrated for the aircraft's speed. The answer is yes, but not in the way you'd calibrate a stationary dish by pointing it once and locking it in place. The Starlink Aviation terminal uses a phased-array antenna. It's a flat panel, roughly the size of a surfboard, mounted on the top of the fuselage. There are no moving parts. No gimbal, no motor, nothing that physically rotates. Instead, it steers the beam electronically by controlling the phase of the signal across hundreds or thousands of tiny antenna elements.
So it's not aiming the way a person aims a satellite dish. It's shaping the beam in software, effectively pointing it by doing math.
And it's doing that math continuously. The satellites in Starlink's constellation are in low Earth orbit, moving at about seven and a half kilometers per second relative to the ground. The aircraft is moving at roughly zero point eight Mach, which is somewhere around two hundred and fifty meters per second, depending on altitude and air temperature. So the relative velocity between the antenna and the satellite it's tracking is not just the satellite's speed or the plane's speed — it's the vector sum of both, and it's changing constantly as the satellite crosses the sky and the plane banks or changes heading.
That's the thing that I think most coverage misses. They talk about the plane moving fast, but the satellite is screaming across the sky at twenty-seven thousand kilometers an hour. The plane's speed is almost a rounding error compared to that. The real challenge is that both objects are moving in different directions, and the antenna has to maintain a lock through all of it.
And it's not just tracking one satellite and holding on until it disappears over the horizon. In a LEO constellation, each satellite is only visible for maybe a few minutes. The antenna has to hand off from one satellite to the next seamlessly, often while the aircraft is in a turn. During a banking maneuver, the antenna's view of the sky tilts dramatically. A mechanical dish would lose lock. A phased array recalculates the beam direction in microseconds and keeps the connection alive.
So the calibration Daniel's asking about isn't a setup procedure. It's a real-time control loop that runs for the entire flight. The antenna is constantly measuring the incoming signal, predicting where the satellite will be in the next few milliseconds, and adjusting the beam. It's also compensating for Doppler shift, because the frequency of the signal changes when the source and receiver are moving toward or away from each other at those speeds.
Right, and that Doppler compensation is built into the modem that sits between the antenna and the network. The Starlink terminal handles all of this internally. The antenna itself has its own processing unit, its own power supply, and its own cooling. On an aircraft like the Emirates A380 — even though Daniel wasn't on one — the terminal is mounted on the fuselage with a radome over it to reduce drag and protect it from the elements. It draws power from the aircraft's electrical system and dissipates heat through the mounting structure.
Which brings us to Daniel's other question about the cellular routers. He noticed that the routers marketed for moving objects are often the same models used for stationary cellular internet, and he's right. The router itself doesn't care whether it's in a building or on a plane. What changes is what's plugged into it.
The key differentiator is external antenna support. A consumer cellular router might have built-in antennas or maybe a single SMA connector for an external antenna. A router designed for a vehicle or an aircraft will typically have multiple antenna connectors — often four or more for MIMO — and it'll support the specific frequency bands and protocols used by the external modem. In the Starlink case, the router inside the cabin isn't doing anything cellular at all. It's connecting to the Starlink terminal via Ethernet, and the terminal is handling the satellite link. The router's job is link aggregation, failover, and distributing the connection to the cabin network.
So the router is the same because the routing problem is the same. NAT, DHCP, traffic shaping, failover between links — none of that changes when you put it on a plane. The part that changes is the modem and antenna upstream of the router.
And the software. In a moving vehicle, you often want to bond multiple connections. A train might have LTE from trackside towers plus a satellite link. A boat might have cellular near the coast and satellite offshore. The router needs to handle seamless failover and sometimes aggregate bandwidth across those links. That's a software feature, and it's why you see routers from companies like Peplink or Cradlepoint marketed for both fixed and mobile use. The hardware is similar, the antenna connectors are more numerous, and the software is configured differently.
Let's talk about what happens once that signal gets inside the cabin, because this is where Daniel's question about AP placement gets interesting. He asked, did they just stick an AP every few dozen rows and hope for the best? The answer is no, but the design constraints are different from what you'd do in an office building.
In an office, you'd do a site survey. You'd walk around with a spectrum analyzer, measure signal strength, identify dead spots, and place APs accordingly. You can't really do that on an aircraft in the same way, because the cabin is a known quantity. It's a long, narrow metal tube with seats in predictable configurations. The RF environment is hostile — the fuselage is a Faraday cage, the seats and overhead bins are full of signal-absorbing material, and you've got three hundred human bodies full of water attenuating the signal.
So the design is more about modeling than surveying. You know the dimensions of the cabin, you know the materials, you know the seat layout, and you simulate the RF propagation. Then you place APs at regular intervals, but the placement isn't random — it's designed to create overlapping coverage cells that hand off cleanly as passengers move through the cabin. Which they mostly don't, but the APs still need to handle the load distribution.
The APs are typically mounted in the ceiling, above the overhead bins, spaced every few rows. On a wide-body aircraft, you might have APs in the center aisle ceiling and along the sidewalls. They're connected back to a network rack in the avionics bay via Ethernet or fiber. That wired backbone runs through the aircraft's cable trays, which are already a nightmare of power, data, and control wiring for the in-flight entertainment system, the cabin lighting, the crew intercom, and everything else.
That's where Daniel's question about the wiring being hidden comes in. It is hidden, but not in the sense that someone artfully concealed it. It's hidden because it's buried in the same cable runs that carry everything else, and those runs are behind panels that require tools to access. On a new aircraft, the Wi-Fi wiring is designed in from the start. On a retrofit, which is what most Starlink installations are, you're threading new cables through existing conduits that were already packed full.
And the avionics integration is real. The Wi-Fi system connects to the aircraft's power bus, it ties into the cabin management system for status monitoring, and in some installations it interfaces with the flight management computer to know the aircraft's position and attitude. That last part matters for the antenna — knowing where the plane is and which way it's pointing helps the phased array predict which satellites will be visible next.
So when Daniel asks whether this is connected to the main avionics, the answer is yes, but not in a way that affects flight controls. It's on the data bus, it gets power from the aircraft, it reports its status to the cabin systems, but it's not talking to the autopilot. There's an air gap between the passenger connectivity network and the flight-critical systems.
The network rack itself — I've seen installations where it's a standard nineteen-inch rack in the avionics bay, with the satellite modem, the router, a switch, and sometimes a server for the captive portal and caching. The whole thing is ruggedized for vibration and temperature, but the architecture is recognizably the same as what you'd find in a small data center. Ethernet switches, fiber transceivers, rack-mount servers. The exotic part is the antenna on the roof.
Let's talk about Daniel's speed test, because a hundred and fifty megabits per second over the Atlantic with low jitter is remarkable. A few years ago, in-flight Wi-Fi meant ground-to-air cellular — towers on the ground pointing up at the planes, with maybe three to ten megabits shared across the whole cabin. And that only worked over land.
The old air-to-ground systems, like Gogo's original service in the US, used cellular towers with upward-facing antennas. The plane had a belly-mounted antenna that connected to whichever tower was in range. Over the ocean, there were no towers, so you either had no connectivity or you used geostationary satellites. Geostationary satellites orbit at about thirty-six thousand kilometers, which gives you a round-trip latency of around six hundred milliseconds just from the speed of light. Fine for email, terrible for anything interactive.
Starlink's satellites are at about five hundred fifty kilometers. The round-trip latency is on the order of twenty to forty milliseconds. That's why Daniel saw minimal ping and jitter. The connection feels responsive because the light doesn't have as far to travel.
And the bandwidth is shared across the satellite's footprint, but each satellite has substantial capacity. The Starlink Aviation service offers plans up to a few hundred megabits per aircraft. Emirates is rolling this out across their fleet, and they've made it free for passengers. No login portal, no time limits, no tiered pricing. You connect and it works.
Which is a business decision as much as a technical one, but it changes the passenger experience completely. When Wi-Fi is metered or expensive, people ration it. They check email and then disconnect. When it's free and fast, they stream video, they make video calls, they do the speed tests that Daniel was doing. The network has to be designed for that load profile.
And the load profile on a plane is weird. You've got three hundred people, most of whom are bored and on their devices simultaneously. But they're not all streaming 4K video. Some are scrolling Instagram, some are reading, some are asleep. The network has to handle bursts of demand — like when the plane levels off and the seatbelt sign goes off and everyone pulls out their phone at the same time.
That's where the AP placement and channel planning matter. You can't just crank up the transmit power on each AP, because then they interfere with each other. You have to manage the channel assignments so adjacent APs aren't on overlapping frequencies, and you have to handle the client roaming so a passenger's phone doesn't cling to an AP three rows back when there's a better one right overhead.
The Wi-Fi standard itself handles a lot of this now. Modern APs support fast roaming protocols, band steering to push clients to 5 GHz or 6 GHz where there's more spectrum, and airtime fairness to prevent one slow client from dragging down the whole cell. But in a metal tube with three hundred clients, the RF environment is still challenging. The reflections off the fuselage create multipath interference, and the density of clients means each AP is handling dozens of associations simultaneously.
Which is why Daniel's hundred and fifty megabits is impressive not just because of the satellite link, but because the cabin network is delivering that to his phone without congestion or interference. The whole chain worked.
Let's broaden this out to other moving objects, because Daniel asked about trains and boats too. Trains have a similar antenna challenge but different infrastructure. A high-speed train in Europe, like a Deutsche Bahn ICE, uses a combination of trackside LTE towers and satellite. The train has external antennas on the roof, often multiple antennas for MIMO, and those connect to routers inside the train that aggregate the links. The train's speed — up to three hundred kilometers an hour — creates the same Doppler and handoff challenges, but the towers are closer together and the handoffs are between ground stations rather than satellites.
The handoff problem on a train is actually harder in some ways, because you're moving past towers every few seconds at top speed, and each handoff has to be nearly instantaneous. On a plane, you're handing off between satellites every few minutes. The frequency of handoffs is lower, but the complexity of each handoff is higher because of the distance and the relative velocities.
Boats have the opposite problem. A cruise ship moves slowly, so the Doppler shift is minimal, but it's on the ocean with no fixed infrastructure anywhere nearby. The only option is satellite. And the ship rolls and pitches, which means the antenna has to compensate for motion in three axes. On a plane, the motion is mostly forward with some banking. On a boat, you've got pitch, roll, and yaw, plus the mast or mounting point is swinging through an arc. Stabilized antennas for marine use have been around for decades — they use mechanical gimbals and motors to keep the dish pointed at the satellite. But phased arrays are starting to replace those too, because they have no moving parts to wear out in the salt spray.
The cellular router observation Daniel made applies to all of these. A boat near the coast might use cellular, a train uses trackside LTE, a plane uses satellite — but the router aggregating those links is fundamentally the same device with different antennas plugged into it. The market for these routers is a subset of the general cellular router market because the core routing function doesn't change. What changes is the number of WAN interfaces, the external antenna support, and the software for bonding and failover.
And that bonding software is where the real differentiation happens. If you're on a train and you've got LTE from three different carriers plus a satellite link, the router needs to decide which packets go where. It might send latency-sensitive traffic like voice over the lowest-latency link, bulk downloads over the highest-bandwidth link, and duplicate critical packets across multiple links for reliability. That's a software problem, and it's the same whether the router is in a data center or on a ferry.
The hardware is a commodity. The antennas and the software are not.
Let's circle back to the Emirates installation specifically, because there's a detail worth highlighting. The Starlink Aviation terminal isn't just a standard Starlink dish bolted to a plane. It's a different piece of hardware, designed and certified for aviation use. It has to meet requirements for temperature, pressure, vibration, and electromagnetic interference that a ground terminal never faces. At altitude, the outside air temperature can be minus fifty Celsius. The terminal has to handle that, plus the thermal shock of descending into a hot, humid airport. It has to survive lightning strikes. It has to not interfere with the aircraft's navigation and communication systems.
The certification process for that is brutal. You can't just stick an antenna on an aircraft and call it a day. Every component has to be tested and approved by aviation regulators — the FAA in the US, EASA in Europe, and whatever the equivalent is in the UAE for Emirates. The installation itself has to be certified for each aircraft type. A Starlink terminal on a Boeing 777 is a different certification than on an A380, even if the terminal is identical.
And the radome — the aerodynamic cover over the antenna — has to be designed for each aircraft type too. It has to be structurally sound at cruising speed, it has to be transparent to the radio frequencies the antenna uses, and it has to not add too much drag. On a long-haul flight, even a small increase in drag translates to measurable additional fuel burn over thousands of hours of operation.
When Daniel says it's remarkable how well the system works from a network design standpoint, he's understating it. It's remarkable from a physics standpoint, an RF engineering standpoint, an aviation certification standpoint, and a network operations standpoint. And the passenger just sees a Wi-Fi icon.
It's the most deceptive little icon in the world.
Before we wrap up the main discussion, I want to touch on something Daniel mentioned about Unifi gear allowing more precision than TP-Link for home setups. He's drawing a parallel to the aircraft scenario, and it's a good one. In a home, you can get away with a single all-in-one router if you don't care about coverage in the far bedroom. On an aircraft, you cannot get away with that. You need multiple APs, wired back to a central switch, with proper channel planning and power management. It's the same principle as a well-designed Unifi deployment, just with more metal in the way and three hundred clients instead of a family of four.
The principle of wiring everything you possibly can and deploying Wi-Fi strategically — that's exactly what's happening in the aircraft cabin. The APs are wired back to the rack. The rack is wired to the satellite modem. The only wireless hop is the last few meters from the ceiling AP to your phone. Everything else is copper or fiber.
That's why the performance holds up. If they tried to mesh the APs wirelessly through the cabin, the metal fuselage would destroy the backhaul links. The wired backbone is not optional.
The answer to Daniel's question about whether they just stuck APs everywhere and hoped is: no, they designed a proper wired network with modeled RF coverage, and the APs are strategically placed to create overlapping cells with clean handoffs. It's enterprise Wi-Fi, just in a tube.
In a very fast tube.
In a very fast tube at the edge of space.
Hilbert: I was a cabin crew trainer for Tower Air back in the nineties. They're gone now. Folded in two thousand.
Tower Air. I remember them. Seven forty-sevens, mostly.
Hilbert: Old ones. Ex-Pan Am frames. The in-flight entertainment was a movie projected on a screen at the front of each cabin section. If you were in row forty, you were watching a postage stamp. And the connectivity — we had a single satellite phone bolted to the bulkhead in business class. It had a handset with a coiled cord, the kind that tangles if you look at it wrong. Ten dollars a minute.
Ten dollars a minute in nineties money.
Hilbert: The crew had to log every call in a paper ledger. Name, seat number, duration, credit card imprint. There was a carbon copy. If the connection dropped, which it did constantly, you had to note the reason in the margin. My favorite was "satellite unavailable." No further explanation.
What did people use it for?
Hilbert: Mostly to tell someone they were on a plane. The call would last forty-five seconds and cost ten dollars. The minimum charge was one minute. Nobody cared. They just wanted to say they were calling from the air.
Now Daniel's running speed tests over the Atlantic and pulling a hundred and fifty megabits.
Hilbert: The thing I've been waiting to say is about the wiring. Everyone talks about the antenna and the satellite and the phased array, and that's all real. But on those old Tower Air planes, the in-flight entertainment wiring was a disaster. It was a retrofit on top of a retrofit. The original Pan Am cable runs were still in there, disconnected, just adding weight. The new system had been installed in three phases by three different contractors, and none of the documentation matched what was actually behind the panels.
When you hear about a modern Starlink install with a clean wired backbone, you're skeptical.
Hilbert: I'm saying the diagram is clean. The reality depends on who did the install and whether the aircraft had three previous systems ripped out. Emirates is probably fine — they run a tight operation. But I've seen cable trays where you couldn't fit a toothpick, let alone another Cat6 run. Some of those planes are thirty years old and they've had five different connectivity systems installed. Each one leaves ghosts.
Ghost cables.
Hilbert: Dead copper that nobody removes because pulling it out would require disassembling half the cabin. So it just sits there. You open a panel and there's a bundle the size of your arm, and half of it goes nowhere.
That's a detail that never makes it into the press release.
Hilbert: The coiled cord on that satellite phone — we had a box of spare cords because they'd fail every few months. The conductors inside would break from being stretched and retwisted. The phone itself was built like a tank, but the cord was the weak point. I still have one of those handsets in a box somewhere. Not sure why.
Because it's a piece of history. Ten dollars a minute to a hundred and fifty megabits for free, in less than thirty years.
Hilbert: The paper ledger is the part I think about. Somewhere there's a warehouse with thousands of those carbon copies. Name, seat, duration, "satellite unavailable." That was the state of the art.
Hilbert, you've just described the entire arc of in-flight connectivity in one image.
Hilbert: I'm just saying the wiring is never as clean as the diagram.
That's a good reminder that the tech is only part of the picture. The operational reality — the install quality, the maintenance, the ghosts of old systems — that's what determines whether the passenger actually gets the hundred and fifty megabits.
It makes you think about where this goes next. The phased-array technology that Starlink uses is getting cheaper and smaller. The same beam-steering approach is showing up in consumer terminals, in marine antennas, in terminals designed for cars and trucks. The moving object problem is becoming the norm. In a few years, the idea that your internet connection should stop working because you're moving will seem archaic.
The cabin network might become the bottleneck before the satellite link does. If the antenna can deliver a gigabit to the aircraft, but the cabin Wi-Fi can only distribute a few hundred megabits across three hundred passengers, that's where the next round of engineering goes. Wi-Fi 7, better channel planning, maybe even millimeter-wave for seat-back connections.
The thing Daniel observed — that a consumer Android phone can hit a hundred and fifty megabits at thirty-five thousand feet — that's the real benchmark. Not what the spec sheet says, not what the airline advertises, but what a passenger with a phone can actually measure. That's infrastructure now. It's not a feature anymore.
We'll see how far it goes when LEO constellations are fully built out and every aircraft has a phased array on the roof. The interesting question is whether the ground infrastructure keeps up, or whether the sky becomes the better-connected place.
Thanks to our producer Hilbert Flumingtop for keeping us honest about the wiring.
This has been My Weird Prompts. If you enjoyed this episode, tell someone about it — or better yet, send us your own weird prompt at show at my weird prompts dot com.
We'll be back soon.
