Offshore monitoring beyond coastal coverage.
A few kilometres past the breakwater the coastal networks give out, and everything beyond runs on a satellite link that charges by the byte. This mission is about making that transition invisible and that budget deliberate: coast and satellite in one bond, sessions that migrate instead of resetting, and a metered link that is spent on purpose — never drained in silence.
Marine researchers, maritime administrations, operators of uncrewed surface vessels on long monitoring missions.
Cellular ends a few kilometres out. Beyond it the only path is satellite — half-second latency, a hard monthly budget, and silent drain.
Coastal cellular and the satellite terminal in one bond; quotas that spend the metered link deliberately; FEC instead of useless retransmissions.
The vessel’s existing onboard computer and the shore workstation. One Linux binary, one configuration file — nothing new afloat.
The mission
An uncrewed surface vessel works a water-quality transect that begins inside the harbour and ends well past the horizon. The shore operations centre wants three things for the whole of it: continuous telemetry and command, the ability to pull up the camera when something on the surface needs a look, and the day’s sensor data in the analysis pipeline without waiting for the vessel to come home. The mission profile is decided by the science and the sea state. It is emphatically not decided by where the coastal carriers happen to end — and that line is crossed twice on every run.
What breaks
Shore cellular at sea is a ragged edge, not a boundary: strong in the approaches, intermittent for a band of kilometres, then gone. Beyond it the only path is a satellite link — and satellite brings two problems of its own. The first is physics: half a second of round-trip latency, which makes retransmission-based recovery of a lossy video stream close to useless, because by the time the repair arrives the moment has passed. The second is economics: the link is metered, and a telemetry stream or a forgotten video session that nobody is watching will quietly eat the month’s budget in days. Operators end up rationing by hand — satellite off unless someone remembers to need it — which converts a connectivity problem into a coverage gap of their own making.
The classical answer, a manual or box-driven switchover between cellular and satellite, fails at the seam: each crossing of the coverage edge resets sessions, drops the command channel for the worst seconds, and still does nothing about the budget.
The architecture on this mission
The vessel runs one node bonding everything it has: the coastal cellular modem and the satellite terminal, in one encrypted tunnel to the operations centre. Every link is probed several times a second, so as the vessel steams outward and cellular thins, traffic migrates to the satellite path before the last cellular packets die — a migration, not a switchover. The tunnel address never changes, so the command session, the telemetry stream and any open video view simply continue across the seam. On the way home the same thing happens in reverse, without anyone touching anything.
The metered link is governed, not rationed by memory. Quotas cap what the satellite path may spend — per class, per period — with the standing rule that a quota never takes down the last usable path: command and telemetry always have their floor. Video over the satellite leg is protected by forward error correction, which repairs packet loss with redundancy sent alongside the stream instead of retransmissions that half a second of latency would make useless. And the bulk of the day’s sensor product does not fight for the metered link at all: it queues on the vessel in priority order and drains for free the moment the bond regains a cellular path on the run home.
How the mission unfolds
- Departure. The vessel powers up alongside; the node brings cellular and satellite into one tunnel and the dashboard shows both links’ measured state. The month’s satellite quota is set in the configuration, not in anyone’s memory.
- Crossing the edge. A few kilometres out, cellular begins to stutter. Weight shifts to the satellite path link by link, before the failing path is gone; the command session and telemetry continue without a reset. The shore operator sees the transition only in the dashboard.
- On station. Beyond the edge, a thin telemetry stream rides the satellite link under quota. When the operator pulls up the camera for a surface contact, the video runs FEC-protected — losses repaired in-line, not re-requested across half a second of latency.
- The squall. Rain fade degrades the satellite path; measured loss climbs. Ordered degradation engages: video sheds quality, telemetry thins toward its floor, and the command channel does not flinch. The quota meanwhile keeps the degraded link’s extra spend visible, not silent.
- The run home. Cellular reappears at the edge and the bond takes it up immediately; the queued sensor data — gigabytes the satellite budget could never carry — drains shoreward in priority order, free. By the berth, the day’s product is already in the pipeline.
- The debrief. The journal holds a per-link record of the whole run: where the edge actually was today, what the satellite link spent and on what, what the cellular path carried for nothing. That record tunes the next mission’s quota and route.
What each mechanism contributes
- Coast-to-sea bonding — cellular and satellite as one measured tunnel; the coverage edge becomes a migration, not a session reset. Shipped.
- Metered-link quotas — the satellite budget enforced in software, per class, with the rule that a quota never kills the last usable path. Shipped.
- Forward error correction — video losses repaired without the retransmission round-trips that satellite latency makes useless. Shipped.
- Class floors and ordered degradation — command and telemetry guaranteed first; video elastic. Shipped, opt-in.
- Store-and-forward with the journal — bulk data waits for the free link; every byte and every link state recorded for the debrief. Shipped.
The honest boundary: Atlas cannot change satellite pricing or satellite physics — beyond the coastal strip the metered link is the only live path, and how much of it to spend is a policy the operator sets, not something software can decide for them. The vessel, its autonomy stack, its marine approvals and the science are yours.
What a pilot should prove
- Session continuity across the coverage edge, outbound and home, with zero re-establishments — from the built-in journal.
- A month’s satellite spend under quota, alongside what the cellular path carried for free on the same missions.
- Video usability over the satellite leg with FEC engaged, against the same stream without it, at measured loss rates.
- A coverage-edge chart of the working area — where cellular actually held on each run — good enough to plan routes and quotas from.
One vessel, one working area, one month of missions. The evaluation format covers the structure.