Skip to content
Contact us

How it works

A sensor network in the Arctic fails for boring reasons. Batteries run flat, a relay dies unnoticed, and nobody finds out until the next field season. The design answer is a radio that knows exactly when to wake up.

01 / The network

Sensor nodes talk to a nearby relay. Relays forward toward a single gateway, which is the only device needing a connection to the outside world. One uplink serves the whole site.

C1 · sensorCarries the instrument. Talks to one parent and never forwards anyone else’s traffic.
C2 · relayForwards across the mesh and acts as cluster master for the sensors around it.
C3 · gatewayThe sink for all data and the source of network time. Backhaul is whatever the site has.
Mesh topology

Three roles, one set of hardware

02 / Why it lasts

Most low-power radios waste their battery waiting for something to arrive. Every node here is given a precise slot, so it wakes, does its business and goes back to deep sleep. Nothing idles with a receiver open.

ScheduledA time-division schedule gives every node its own wake, transmit and sleep window.
Guard bandsEach slot is padded for clock drift, measured across temperature rather than assumed.
Deep sleepThe sensor rail is switched off between samples, so an attached instrument draws nothing.
Slot schedule
one frameC1 · aC1 · bC2C3one slot, magnifiedtxrxguardguardradio statedeep sleepguardtransmitreceiveradio on for roughly 100 ms

One reporting cycle, radio on for roughly 100 ms

03 / When something breaks

Every node keeps a list of the other relays it has heard from. When its parent stops beaconing, it promotes one of them and re-registers. There is no manual topology file and no site visit.

BeaconsRelays announce their identity, position in the tree and routing cost.
Power awareRouting cost is not hop count. A relay whose battery is falling advertises a rising cost, so traffic moves to healthier nodes.
Ages togetherBecause load shifts away from depleting relays, the network degrades evenly instead of losing one hotspot at a time.
Parent failure
beacons missedneighbourscostC2 · Agone quietparentC2 · Bnew parentstandbyC2 · Cstandby

A failed hop is replaced from the neighbour list

04 / Power-aware routing

A relay advertises what it costs to send through it, and that cost rises as its battery falls. When a relay runs low, the nodes around it send their readings another way. Once it has recharged, it is the cheapest route again and the traffic comes back.

BeaconsEvery beacon carries the relay’s current routing cost, so neighbours choose on up-to-date numbers.
BatteryA falling battery raises the cost a relay advertises. Nothing is switched off by hand.
RecoveryA relay with a solar input recharges between reports, then rejoins the cheapest route on its own.
Power-aware routing
route costvia relay 1via relays 2, 3relay 1 battery

A relay runs low, traffic moves away, then returns once it has charged

05 / Pricing a hop

Routing cost is not hop count. Each link is priced from how far the signal has to travel, how much battery the receiving relay has left, and how many hops remain to the gateway. A node adds up the price of each route it could take and sends along the cheapest.

DistanceA longer, weaker link costs more than a short, strong one.
BatteryA relay with less energy in reserve charges more to carry traffic.
Hops leftEvery extra hop to the gateway adds to the price, so shorter routes are preferred.
Pricing a hop
route pricevia Avia Bcheapest

The nearer relay is low on battery, so the route through the other is cheaper

06 / Joining

  1. Discovery

    It listens for relay beacons, picks a parent, and keeps every other relay it heard as a fallback.

  2. Join and slot assignment

    It registers with that parent and is given a slot. From here the radio schedule is fully determined, so there is no reason to open a receiver at any other time.

  3. Synchronised operation

    The gateway is the time anchor, and time propagates down the tree. Slots stay aligned as the hardware warms and cools.

Joining the network
heardR1R2parentframeR1R2C1new

07 / The hardware

A C1, a C2 and a C3 are the same hardware in different configurations, so a deployment is one part number to build, stock and repair. Instruments connect over the interfaces operators already have, rather than requiring a new capital base of probes.

The assembled V1 board, green, with terminal blocks along its edges and an antenna cable attached
The V1 board, assembled
Close-up of the antenna connector and the 868 MHz radio section of the V1 board, with a spectrum trace on an instrument behind it
The 868 MHz radio section during RF testing
The bare V1 circuit board before assembly, showing gold pads and silkscreen labels for the battery, sensor and power sections
The bare V1 board before assembly

  • Validated in a climate chamber to -40°C.

  • One board set, configured as sensor, relay or gateway.

  • 868 MHz, license-free across the EU and Greenland.

Three roles, compared

C1

Sensor

Carries the instrument.

TrafficSends its own readings and never forwards anyone else’s
Place in the treeTalks to one parent
Network timeTakes time from its parent
UplinkNot applicable
C2

Relay

Forwards across the mesh.

TrafficForwards toward the gateway
Place in the treeCluster master for the sensors around it
Network timePasses time down the tree
UplinkNot applicable
C3

Gateway

The sink for all data.

TrafficReceives every reading on the site
Place in the treeThe single way out
Network timeThe time anchor for the whole network
UplinkWhatever the site has
HardwareThe same board set in all three roles

Inside one node

In each reporting cycle
radio on for roughly 100 milliseconds
Enforced in firmware, not left to the operator
a duty cycle of 1 percent
ETSI M-band, license-free in the EU and Greenland
868 megahertz
Measured across temperature in the chamber
0.03 parts per million

Shared by all three

Compute and radioSTM32WL55, application core and sub-GHz transceiver on one die
Sensor portsSDI-12, RS-485, two analog inputs and a switched excitation channel, on Phoenix Contact headers that take more than 100 connector types
PowerSolar MPPT or USB-PD, with a 4S Li-ion pack and a power-gated sensor rail

08 / Honestly

The hardware has been validated in a climate chamber to -40°C, where over 98% of about 19,600 packets were acknowledged and the crystal clock held to 0.03 ppm. That is a measurement.

Field life, coverage and range are modelled from component behaviour and have not been observed over a season in the field. We publish them as design targets, marked as such, and will replace each one with a measurement. The first field deployment is planned for summer 2027.

Every figure, in one place

Measured values and V2 design targets, side by side and labelled.

See them on Why Nuna