Specific Emitter Identification · Iridium NEXT Constellation
The Radio Accent of Satellites
A radio message can carry a forged identifier. Its waveform may offer additional clues about the transmitter, but those clues must hold across dates and environments. IriSig-Spoof compares these conditions using 5.17 million messages from 66 Iridium satellites. Here you can distinguish a signal illustration from the study’s measured results.
What a waveform can reveal
Analog components and propagation can alter the received signal. A useful radio fingerprint must be distinguishable from the effects of rooms, obstacles and reception. The drawing below shows how a few parameters can change an I/Q point cloud.
All five examples are authored illustrations. The A–D labels, groups, offsets and spread do not describe measured satellites. Selecting a preset does not run a classifier or authenticate a signal.
IriSig-Spoof · source and experiment
The illustration uses authored parameters; measured study results are available in the table below.
What the terrestrial transmitter tested
The real generated messages used SAT 1, an identifier not assigned to an operational satellite. The authors thereby avoided impersonating an existing satellite. Section III-C of the paper describes the message format and other fields.
In a separate test, the Legacy model achieved AUROC 0.9715: a ranking measure across decision thresholds for known and unknown signals. At the threshold calibrated on legitimate validation signals, the unknown rejection rate was 0.8583. Neither number is the probability that the selected example below is forged.
IriSig-Spoof · source and experiment
Recognising the radio or its surroundings?
Figure 11 compares a MACNN binary detector across eight test configurations. 1S uses an office with a direct transmitter–receiver path; 2S adds the office without a direct path; 3S adds an office corridor. None of these training variants includes outdoors.
Choose the number of training configurations and signal preprocessing. The results below are transcribed from the legends of the six original plots. AUROC below 0.5 means reversed score ordering, not a percentage of correctly classified messages.
Training on 3S improves transfer in this experiment. The comparison is consistent with propagation-dependent shortcuts; it does not establish that every model learned only the room or that diverse training guarantees robustness everywhere.
IriSig-Spoof · source and experiment
| Test | 1S · No preprocessing | 2S · No preprocessing | 3S · No preprocessing | 1S · Standard preprocessing | 2S · Standard preprocessing | 3S · Standard preprocessing |
|---|---|---|---|---|---|---|
| Office · direct path | 1.0000 | 1.0000 | 1.0000 | 0.9994 | 0.9986 | 0.9980 |
| Office · no direct path | 1.0000 | 1.0000 | 1.0000 | 0.9996 | 0.9998 | 0.9988 |
| Office corridor | 0.5533 | 0.9994 | 1.0000 | 0.4212 | 0.3440 | 0.9979 |
| Atrium balcony | 0.9990 | 1.0000 | 1.0000 | 0.9999 | 0.9999 | 0.9999 |
| Atrium balcony · no direct path | 0.0602 | 0.1564 | 0.8798 | 0.3262 | 0.4126 | 0.9974 |
| Atrium main lobby | 1.0000 | 1.0000 | 1.0000 | 0.0751 | 0.3156 | 0.9957 |
| Outdoors · 25 m | 0.0965 | 0.4417 | 0.9963 | 0.6133 | 0.5006 | 0.9935 |
| Outdoors · 50 m | 0.1144 | 0.7854 | 0.9992 | 0.9944 | 0.9919 | 0.9837 |
What still needs testing
A strong benchmark result does not certify a receiver deployed under other conditions. The study distinguishes identity assignment from unknown-signal rejection and keeps cryptography essential for message authentication.
The paper does not demonstrate a compromise of an encrypted military channel. This page receives no radio signals, runs no research classifier and cannot evaluate a real transmitter.
Sources and limits
Results come from IriSig-Spoof, arXiv version 2608.18642v1, dated 19 August 2026. Reception covered 32 days, 27 June–28 July 2025. The temporal test uses seven days for training/validation and 25 later days for evaluation.
The 97.75% value is the best average cross-day identification accuracy among the compared configurations. Table IV describes the unknown-signal test; Figure 11 describes the binary detector trained on propagation configurations. These are different experiments.