As part of the Svalbard field campaign, we tested and validated our live update framework and communication setup for autonomous, intelligent sensors operating in Arctic conditions.
๐ง ๐๐ข๐ฏ๐ ๐๐๐ฎ๐ซ๐๐ฅ ๐๐๐ญ๐ฐ๐จ๐ซ๐ค ๐๐ฉ๐๐๐ญ๐
A key goal was to verify that our low-power STM32-based sensor device could receive, authenticate, and install updated neural-network weights while running in the field.
The test worked as planned: the device received the update, validated it, applied new weights, and continued live inference without interruption.
This confirms that our architecture for long-term autonomous monitoring is reliable even in harsh outdoor conditions.
๐ก ๐๐๐ ๐ฆ ๐๐๐ง๐ ๐ & ๐๐ฉ๐๐๐ ๐๐๐ฌ๐ญ
We then evaluated the communication link by walking the STM32 node away from the Raspberry Pi hotspot (with UNISโs Geert Hensgens providing polar-bear safety).
The results were strong:
ยทย ย ย ย ย ย ย ย Stable Wi-Fi connection up to 250 m
ยทย ย ย ย ย ย ย ย Upload speed around 9.5 kbit/s, close to the 11.5 kbit/s theoretical maximum
These tests show that the system can reliably support remote neural-network updates over practical Arctic distances.
๐ ๐๐ซ๐จ๐ง๐ ๐๐จ๐ญ๐ฌ๐ฉ๐จ๐ญ โ ๐๐๐ฌ๐ฌ๐จ๐ง๐ฌ ๐๐๐๐ซ๐ง๐๐
We also mounted the Raspberry Pi hotspot on the drone to test airborne communication.
Here, the usable range dropped to less than ~15 m, giving us important input for the next development steps.
Two likely causes emerged:
1๏ธโฃ Antenna orientation โ the current setup is optimized for horizontal, ground-level communication.
2๏ธโฃ Power interference โ powering the Raspberry Pi from the drone battery may introduce electrical noise; a powerbank was tested as an alternative but interfered with the droneโs compass.
The photos show some impressions from the range tests around the station and from the drone-mounted hotspot setup. Together, these results give us a clear direction for improvement: antenna optimization and better power isolation for UAV-based deployments.
๐ธ More insights from the Svalbard campaign coming soon!
