An Iridium receiver can decode a message during radio jamming and still report that the signal stood out clearly from the background noise. In an experimental study, researchers found signs of interference in the radio samples behind those readable messages. That gave them a way to detect jamming before the satellite link stopped carrying usable data.
Ingrid Huso, Giulia Fliescher, Savio Sciancalepore, and Gabriele Oligeri described their experiments in a November 5, 2026, preprint. The team collected Iridium signals under 47 jamming conditions. The tests covered two environments, stationary and moving receivers, and changes in the jammer’s distance, transmit power, and signal type.
What a Readable Message Reveals
A receiver that examines only correctly decoded messages sees a selected portion of the radio traffic. Interference may already have spoiled weaker transmissions. The messages that survive can still have a high signal-to-noise ratio, a measure of how strongly a transmission stands out from the noise around it. Two conventional indicators of signal quality performed poorly as jamming detectors when applied to those surviving messages.
The researchers tested three other detectors that learned patterns in normal radio reception. Each used an autoencoder, a model trained to reconstruct its input. The detector compared recorded radio samples with the model’s reconstruction and used the difference to score how unusual the recording was. It examined the radio signal itself rather than the information carried in the message.
The change in the receiver’s background noise level, called its noise floor, helped explain when the method worked. When a noise jammer raised that level by at least 9 decibels, all three detectors scored at least 0.84 on a measure called area under the curve. A score of 0.5 corresponds to chance when ranking a jammed example against a normal one; a score of one means the detector consistently gives the jammed example the higher score. The reported 0.84 measures that ability to distinguish the two conditions, not the percentage of individual messages classified correctly.
Each test input contained 1,000 samples, equivalent to 40 milliseconds of recorded signal or about four correctly decoded messages. The 40 milliseconds measures the amount of signal analyzed. The elapsed time a terminal takes to collect four messages is a separate quantity. The authors also released the recordings so other detectors can be tested against the same interference conditions.
