Border surveillance across coverage gaps.

The mission

A mountain border sector, tens of kilometres long, patrolled from a station at its edge. The aircraft flies the line several times a day — thermal at dawn and dusk, zoom by day — and its feed drives the response: where the sensor operators see movement, a ground team is tasked. The terrain that makes the sector hard to walk is the same terrain that makes it worth patrolling from the air — and the same terrain that breaks the link.

What breaks

The classical setup is one directional link from the station, and it fails in the classical ways: the aircraft descends behind a ridge to look at something — gone; it reaches the far end of the sector — marginal; weather sits on the middle third — gone again. Long stretches have no cellular coverage of any kind, so there is no fallback to fail to. Every link loss is a procedure — climb, hold, return — and a procedure executed thirty kilometres out is a patrol that did not finish. The deeper risk is sharper: an aircraft that loses its command channel far from home, over terrain, is an aircraft the force may simply lose.

The architecture on this mission

Three moves, composable and independent. First, a relay node on high ground — a mast at a peak the sector already uses for other equipment — joins the mesh and splits the long, fragile path into two short, strong hops. The aircraft’s traffic routes through it automatically when the direct path degrades and back when it recovers; each hop is separately measured and separately encrypted.

Second, a narrowband serial radio — 900 MHz, tens of kilobits — rides in the bond as the line of life. The class system keeps video off it by declared capacity alone, and the command channel’s reserved floor fits comfortably in what it offers: the mechanisms treat tiny links as first-class citizens, proven down to 64 kbps serial radios. If every wideband path is gone — terrain, weather, failure — encrypted command and position continue on a channel that shares no frequency, no modulation and no physics with what was lost.

Third, wherever the sector does have cellular, those modems simply join the bond: used when present, ignored when absent, never depended upon. The patrol’s communications plan stops being a bet on any single layer.

How the patrol unfolds

1. Dawn launch. The station wakes the aircraft for the thermal window. Its node brings up the directional link and the serial radio and bonds them; the mast relay on the first ridge is already meshed in, and every hop is probed several times a second.
2. Out along the line. The aircraft works the near third of the sector on the direct path. The class system keeps video off the serial channel by declared capacity alone; the command floor fits comfortably everywhere.
3. Behind the ridge. The sensor operators see movement and the aircraft descends to look. The direct path dies against the terrain — and traffic is already routing through the mast, two short hops in place of one long marginal one. The zoom feed continues; nobody executes a lost-link procedure.
4. Weather on the middle third. Rain sits on the sector and every wideband path degrades together. Encrypted command and position continue on the 64 kbps serial radio — a channel sharing no frequency, no modulation and no physics with what was lost. The patrol continues; the video waits.
5. Recovery. Climbing out of the weather, the wideband paths return and video re-joins on its own. Near the border village a cellular pocket appears, joins the bond, and is gone again ten minutes later — used, never depended upon.
6. The debrief. The sector link record shows where each path held across the patrol line — the evidence from which the next relay mast’s position gets argued.

What each mechanism contributes

• Mesh relaying with per-hop encryption — the mast as an authenticated member node; routes re-form around geometry without operator action. Shipped.
• Tiny-capacity links as first-class citizens — declared capacity keeps bulk traffic off the serial radio automatically; the command floor fits in kilobits. Shipped.
• Class-ordered degradation — video yields first, position and command last; the order is enforced, not hoped for. Shipped, opt-in.
• Continuous measurement and the journal — a per-sortie record of where each link held across the sector, feeding the placement of the next relay mast. Shipped.

The honest boundary: Atlas contributes the network layer. The aircraft, the masts, their power and the patrol doctrine are the force’s. And a plain statement, because this page will be read carefully: the mechanisms here address terrain, distance and single-link fragility, and they degrade in an ordered way whatever the cause of loss — but resilience against deliberate interference is not a field-validated claim we make. Physical-layer measures remain the counterpart for that threat, and the architecture multiplies them rather than replacing them.

What a pilot should prove

• Command and telemetry continuity across the defined shadow zones of one real sector, with the direct link demonstrably failing there.
• The line-of-life drill: wideband paths disabled in flight, encrypted command and position continuing on the serial radio alone, video re-joining cleanly on restoration.
• Relay handover: routes re-forming without operator action as the aircraft transitions between direct and relayed geometry.
• A sector link record from the built-in measurement, good enough to argue the next mast’s position from.

The evaluation format covers the structure of exactly this kind of trial.