Coastal search and man-overboard response.

The mission

A ferry reports a man overboard forty minutes after the fact. The drift model puts the datum eight kilometres out and moving; the coordination centre launches two lifeboats and a spotter aircraft. What the search needs from its communications is one shared, current picture — the boats’ tracks, the swept areas, the aircraft’s camera, every sighting the moment it happens — because a search pattern is only as good as everyone’s knowledge of what has already been searched. The datum does not care that it sits astride the exact line where shore networks give out.

What breaks

Coastal cellular is built to serve the land and merely leaks out to sea. A few kilometres offshore it thins, then dies — and the boundary is not a line on a chart. It moves with sea state, with a masthead antenna’s height above the swell, with the boat’s heading as its superstructure shadows the antenna. One boat holds a usable link while its sister ship, two kilometres further out, holds nothing. Marine VHF still works, but it carries voice — not tracks, not video, not the sweep map. So the picture fractures precisely along the coverage edge: the coordination centre sees the near units, hears the far ones, and reconstructs the rest by radio discipline and guesswork.

The classical answers are unsatisfying. A satellite terminal on every lifeboat is cost and antenna real estate that small fast boats do not have. Keeping the search inside coverage is not an answer at all — the drift model, not the coverage map, decides where the person is.

The architecture on this mission

Every unit runs the same software node: the shore station, both boats, the aircraft. Together they form one encrypted mesh, and the load-bearing move is relaying across the coverage edge: a unit still inside shore cellular carries the traffic of the units beyond it. The far boat’s short hop to its sister ship survives geometry that kills its long hop to shore; the aircraft, high over the datum, is both a camera and a relay that sees far past what any masthead antenna can. As boats cross the edge in and out — and they do, continuously, because the pattern demands it — every link is probed several times a second and traffic re-routes with no switchover event and no operator action.

Where the boats also carry marine broadband radio, those IP links join the same bond as shore cellular: two dissimilar paths whose bad moments do not coincide. Traffic is classed on top — tracks, tasking and voice hold reserved floors; the aircraft’s video is elastic and sheds quality first when a hop narrows. And a sighting is treated as what it is: the single most valuable datagram of the day. Detections and geotagged frames travel store-carry-forward, queuing through any gap between hops and delivering on the next contact, in priority order. A gap delays a sighting by seconds; it can no longer delete it.

How the callout unfolds

1. The call. The MOB report reaches the coordination centre; the station node is already up. The boat and aircraft nodes authenticate by key and join the mesh as engines start — no configuration on the slipway.
2. Transit out. The near boat holds shore cellular; the far boat crosses the edge inside the first twenty minutes and its traffic re-routes through its sister ship automatically. On the shore screen, nothing changes — both tracks keep painting.
3. The pattern. The aircraft orbits the datum and its camera feeds the coordination centre over whichever path currently holds. Tracks, sweep status and tasking rank above video; the sweep map stays current for every unit at once.
4. The sighting. A thermal flash in the swell, caught while the aircraft is momentarily between hops. The detection queues on board, delivers seconds later over the next contact, geotagged and time-stamped — and the nearest boat is vectored onto it.
5. The recovery. Everything converges on one position. The vectoring channel holds its reserved floor while video sheds quality into the congestion; the command never contends with the camera.
6. The debrief. The journal holds measured coverage along the actual search tracks — where the edge really ran that day, which hops carried what. That artefact plans relay positioning for the next callout and gives the station a coverage map built from missions, not predictions.

What each mechanism contributes

• Mesh relaying across the coverage edge — units inside cellular carry units beyond it; routes re-form as boats cross the edge, with per-hop encryption throughout. Shipped.
• Bonding shore cellular with marine IP radio — two dissimilar paths as one measured connection; the search stops depending on either alone. Shipped.
• Traffic classes — tracks, voice and vectoring reserved; video elastic; the sweep map never waits behind the camera. Shipped, opt-in.
• Store-carry-forward — detections and geotagged frames queue through gaps and deliver on the next contact, priority first. Shipped, opt-in.
• Continuous measurement — a per-link record along every track: where the edge actually is, mission after mission. Shipped.

The honest boundary: Atlas cannot move the coverage edge — a boat beyond every fragment and every relay is genuinely alone until the geometry changes, and it is the operation’s job to keep a relay unit where the pattern needs one. The boats, the aircraft, the drift models and the seamanship are the crew’s. What the mechanisms demonstrably survive is the fading, shadowing and dropout that a moving search across a coverage boundary is actually made of.

What a pilot should prove

• Measured coverage along real search tracks, with and without relaying through the near unit, from the built-in per-link record.
• Picture continuity for a boat working beyond the edge: track, tasking and sweep status unbroken while its direct shore link is measurably dead.
• Detection delivery: sightings generated during induced gaps arrive complete, geotagged and in priority order after reconnection.
• A debrief artefact: the journal’s coverage map of the exercise area, usable for the station’s relay doctrine and the next callout.

Two boats, one aircraft, one exercise. The evaluation format covers the structure.