Defense

FMV and telemetry over a contested link

On a thin tactical link the downlink is the binding constraint. The video stream is already near-optimal under H.264 or H.265, so Polygen leaves it alone and tokenizes the telemetry and per-frame signals beside it: about 200 bytes per minute of telemetry, 4 bytes per frame of event signal, and analytics on tokens with no decode.
Business Impact

What changes when raw data stops moving

Reclaim the telemetry and metadata channels on a fixed downlink
Detect events at the edge without sending video
Run analytics on tokens under size, weight and power limits
Verify mission data bit-identical at the ground station

125x

Smaller telemetry, coefficients only
21-channel log, about 200 bytes per minute; lossless 5x, gzip 2x

4 B

Per frame for event detection
Scene-cut and event signal from the codec residual

~50x

Faster per-channel mean on tokens
200 B read against 25 KB, one-minute window

92%

Less transmitted than full-stream x265
Event-gated ROI on a public drone mission, gate fixed before measurement
The challenge

Every byte transmitted and every cycle spent decoding costs power

Drone-borne capture plus ground-station analytics across a heterogeneous fleet and an intermittent network. Bandwidth back to the ground station and on-device analytics under size, weight and power are the two pressure points. Polygen is not a video codec; it compresses the structured signals the codec does not address.
Approach

How Datasent enables this use case

Encode

Pinned basis, no search in flight

Telemetry is canonicalised to integers and tokenised against a pinned basis on the platform. Adaptive segmentation fits flight phases, and the per-segment residual scale flags manoeuvres and sensor faults at the edge.
Transmit

Coefficients over the link

Ship the coefficient form: roughly 504 bytes per segment against 8,400 raw. Suppressed video frames send a 4-byte energy value; event frames send region-of-interest tokens.
Reconstruct

Exact segments for forensics

The ground station decodes only the segments an alert points at, losslessly, and verifies them against the per-segment digest. The rest of the archive stays encoded.