Edge Computing Architecture in Remote Tower Systems

Edge computing keeps remote tower systems safe by processing data locally instead of the cloud.

· 10 min read · Updated
Cover illustration for “Edge Computing Architecture in Remote Tower Systems”
Remote Tower Architecture · October 1, 2026 · 10 min read · 2,290 words

Most airports in the United States run without any air traffic control, and that gap exists because the economics of a traditional tower have never scaled down to fit a small field, not because anyone decided the safety case didn't warrant it. The country has roughly 500 towered airports set against tens of thousands of non-towered landing facilities, and that imbalance is built into the system rather than a problem the system forgot to fix. A single-position traditional control tower costs several million dollars, and that's before you add the cost of staffing it around the clock or maintaining the equipment, so if you're a small general-aviation airport running on a modest federal runway-extension grant, that number simply isn't reachable.

Three pressures are converging to make this arrangement harder to hold together. The country faces a controller shortage of more than 3,500 positions, the physical tower infrastructure that does exist is aging, and staffing a tower at an airport with low traffic volume costs more than the traffic can justify. The capital picture is tightening at the same time the need is growing: once FY 2026 closes out, the largest supplemental federal airport capital program is exhausted, with its final selections announced May 18, 2026, leaving AIP entitlements and discretionary grants as the primary federal capital channels for non-hub airports. Fewer dollars are chasing a wider set of infrastructure needs.

At an airport with no tower, safety rests on pilots separating themselves from each other voluntarily. That means staying alert, looking out the window, and exchanging traffic information over the radio, a task complicated by the fact that some aircraft using these fields don't carry two-way radios at all. This is not a flaw confined to small fields. Visual separation as a backstop is a feature of the whole air traffic system, and a towered airport is not automatically exempt from its limits. The January 2025 DCA collision, whose NTSB investigation concluded with a board vote in January 2026 and a final report released in February 2026, showed that even at a fully staffed, high-traffic towered airport, the system relied heavily on pilot-applied visual separation without adequate consideration of the inherent limitations of see-and-avoid.

Airport infrastructure itself plays a measurable role in accident outcomes. The IATA 2025 Annual Safety Report found that airport facilities contributed to a meaningful share of 2025 accidents, which puts runway markings, lighting, and surface conditions in the same category as any other active cause of accidents rather than a detail in the background. A funding structure can't put a traditional tower at every airport that could use one, and a safety doctrine leans on human vision even where towers exist, so what kind of coverage can close the gap without a several-million-dollar building and a full staffing roster at every field? That's the question a remote tower is built to answer.

What a remote tower must accomplish

A remote tower is not a camera bolted to a pole and streamed to a monitor somewhere else. It rebuilds, in full, the view a controller would have looking out a tower window, along with all the real-time data integration that air traffic control depends on, and building that reconstruction accurately is the hard engineering problem at the center of the whole technology. The system has two core pieces: sensors at the airport itself, and controller working positions at a remote control centre that may sit anywhere from right next to the airfield to thousands of kilometers away, connected by a secure data network.

The sensor side includes high-definition camera arrays that cover a panoramic field of view, infrared sensors for night and low-visibility conditions, and zoom capability for closer inspection of specific aircraft or runway sections. Video from those cameras combines with radar and ADS-B tracking data into a single augmented-reality display at the controller's station. That display has to carry more than a live picture. It needs aircraft identifiers tagged onto the video feed, system status alerts, the outlines of runways and taxiways overlaid on the image, and zoomed-in views the controller can call up on demand, all presented as one continuous and natural scene rather than a patchwork of separate windows.

Latency is the hard number that constrains all of this: to feel real to a controller making split-second calls, the reconstructed image of the airport has to be captured and displayed in under one second in the worst case, and that single constraint shapes nearly every decision an engineer makes about how to build the rest of the system. A remote tower can run in single mode, where one controller handles one airport, or, with enough automation built in, in multiple mode, where one controller oversees several airports from the same position. Multiple mode has not reached operational deployment anywhere yet and remains in research and development. The role a remote tower plays can also vary: it can be the primary air traffic control facility for an airport that never had a tower, or it can serve as a backup, including covering a Class D airport overnight when the physical tower would otherwise shut down for the night.

Artificial intelligence is now layered onto that sensor and display foundation. Image-processing systems analyze the high-definition camera feeds as they come in, automatically flagging aircraft movement, tracking potential runway incursions, and picking out weather hazards before a controller might spot them by eye. These AI engines have been run against large flight volumes at major airports and have recovered meaningful amounts of capacity that would otherwise be lost when visibility drops. None of that, the cameras, the overlay, the AI detection, works if the underlying data can't move fast enough to stay under the one-second threshold. That's the constraint that determines where the processing has to happen.

Why cloud-only processing fails the one-second requirement

Routing live runway video and ADS-B data to a cloud server for processing, then waiting for the results to come back, isn't a design choice that can be tuned until it's fast enough. When a cloud architecture introduces round-trip delays like that, it breaks the basic safety premise air traffic control is built on. Cloud-based AI systems carry three failure modes that matter enormously here: high end-to-end latency, dependence on a connection staying up the whole time, and exposure to privacy and security breaches. In most computing applications, those are inconveniences. In an ATC context, where the data is safety-critical and the window to act is measured in seconds, they're disqualifying.

You send sensor data out to a cloud server and wait for an inference result back, but that can simply take too long for what flight operations require, and remote airports can't always count on the high-bandwidth, low-latency connections that kind of round trip needs. When that link degrades, a cloud-dependent system has almost no ability to keep working offline. Bandwidth demand compounds fast. A single high-definition camera stream already asks a lot of a network connection. A panoramic array covering every runway and apron angle multiplies that load, and once radar, ADS-B, weather sensor data, and voice communications are added on top, the result is a continuous volume of data that a constrained rural or remote backhaul connection can't carry reliably.

The airports that need remote towers the most are also the ones least likely to have a redundant, high-bandwidth fiber line running to a distant data center. That's the central practical obstacle: coverage is needed exactly where the infrastructure to support a cloud round trip doesn't exist. Aviation data carries its own sensitivity. Moving large volumes of raw video and voice traffic to an external cloud multiplies the points where that data could leak or run into regulatory trouble, and every additional network hop between the airport and the processing server adds to the cybersecurity exposure of a live ATC feed.

Cloud latency has improved over the past several years, and connectivity infrastructure keeps expanding into rural areas. Neither trend changes the math here, because the one-second requirement leaves no margin for a round trip, and the airports most in need of a remote tower sit outside the urban fiber corridors where connectivity has actually gotten better. The architecture has to be built for the worst realistic connection a small airport will have. Once that constraint is accepted, moving the processing closer to the airport stops looking like an optimization and starts looking like the only workable answer.

Edge computing at the airport node

Edge computing puts the latency-sensitive work, video analysis, object tracking, conflict detection, speech recognition, on hardware located at or near the airport itself. Only compressed, already-processed output travels over the network to the remote controller. Edge intelligence means building AI capability into end devices, embedded processors, and local servers at the network's edge instead of routing everything through a centralized cloud, and running the models on those distributed nodes cuts communication latency dramatically.

In a remote tower system, that translates to a specific data flow. The camera array, radar feed, and ADS-B receiver at the airport all connect first to a local edge server, typically sitting on an equipment rack or mast at the airport site. That server does the heavy computational work, stitching together the panoramic camera images, detecting and tracking objects on the airfield, generating the augmented-reality overlay, before any of it leaves the property. What goes out over the link to the remote control centre is a finished, composed view with the tracking data already computed. That single change drops the bandwidth requirement substantially and takes the latency budget out of the hands of raw data transit.

Voice communications get the same treatment. Speech recognition systems tuned for noisy VHF radio traffic run on the edge node itself, so they catch call-sign confusion or readback mistakes in real time, and they don't wait on a round trip to a cloud server. The cloud still has a role, improving the underlying language models between shifts, but the task that can't wait, catching an error as it happens, stays local. That split, local inference for anything time-sensitive and cloud support for background improvement, is the practical shape the edge-plus-cloud model takes in this setting.

Keeping the data local solves the security exposure as well as the latency constraint. Raw runway video never has to leave the airport site, so it cuts down the bandwidth a wide-area link needs to carry and shrinks the number of places a cyber intrusion could target a live ATC feed. The architecture also degrades gracefully. If the link to the remote control centre drops or slows down, the local edge system keeps capturing and logging data on-site, and once the connection comes back, that gap in the record can be reconciled instead of simply lost. That resilience is a property of how the system is built.

Faster transport layers like 5G and satellite communication matter here, but mainly because once the heavy processing is already done, they just make the final hop quicker. Fast transport alone would still leave raw, unprocessed data moving across the link. The edge layer is what makes 5G or satellite transport sufficient for the task, not just fast in the abstract.

What operational deployments show about the edge architecture

The deployments furthest along share the same basic architecture: processing done locally at each airport, a secure wide-area network linking that airport to a centralized controller facility, and AI running at the edge instead of waiting on a cloud response.

London City Airport is the clearest example of this working at full scale. In January 2021, it became the first major international airport anywhere to be fully controlled by a remote digital ATC tower. The airport was in the middle of a major redevelopment at the time, and moving ATC off-site freed up airside space the redevelopment needed. Separately, London Heathrow has been testing a Digital Tower Laboratory: it's built around 20 ultra-HD cameras that feed into an AI platform called AIMEE. That system has been run against tens of thousands of flights, alerting controllers when aircraft have cleared a runway and recovering capacity that would otherwise be lost to poor visibility.

The technology itself has been proven elsewhere, but in the United States, the regulatory progress has lagged far behind. The FAA Reauthorization Act of 2024 required the FAA to build a clearly defined system design and operational approval process for remote and digital towers and to publish milestones for testing and deployment, but the agency has not yet published a system design approval for any project. That absence matters because it leaves airports and vendors without a clear regulatory path to certify a remote tower system for use, even where the underlying technology has already been proven at places like London City and Heathrow.

Industry coordination is building around the gap in the meantime. The Digital Tower Technology Coalition belongs to the Modern Skies Coalition, which formed in 2025 to back the U.S. Department of Transportation's multibillion-dollar push to modernize air traffic control. On the funding side, the FAA's Contract Tower Competitive Grant Program has distributed money every year from FY 2022 through FY 2026, with up to an additional $100 million in unobligated AIG funds added in FY 2026, under the Bipartisan Infrastructure Law, for tower modernization work that includes remote tower certification.

The regulatory timelines differ sharply, but the pattern across both markets is the same. Wherever a remote tower has moved from pilot project to full operational control, the system doing the heavy lifting sits at the airport, not in a distant data center. The U.S. has the funding mechanisms and an industry coalition in place, and is still waiting on the FAA to publish the approval pathway that would let a comparable project move from testing to full certification.

Sources

  1. Edge Intelligence in Civil Aviation: Paradigms, Techniques, and Applications
  2. Coalition Pushes for Remote and Digital Air Traffic Control Towers
  3. Remote Towers Market Developments: Virtual ATC, 360-Degree Vision & Market Expansion to 2033
  4. Digital Towers: The Future of Air Traffic Control
  5. Digital and Remote Towers: a new paradigm for aerodrome control service
  6. A solid year of safety performance - Airlines Magazine - IATA

More in Remote Tower Architecture