Skip to content
Marius Comper

Specific Emitter Identification · Iridium NEXT Constellation

The Radio Accent of Satellites

A received data packet can declare whatever identity it chooses within its digital protocol fields. Yet the physical electromagnetic waveform carries something distinct: microscopic imperfections produced by the amplifiers, filters, and local oscillators of each individual spacecraft. Analyzing 5.17 million Ring Alert transmissions collected continuously across 32 consecutive days, researchers demonstrated that all 66 operational Iridium satellites possess a persistent hardware „accent”, enabling receivers to reject a terrestrial SDR spoofer even when it transmits perfectly constructed, protocol-compliant packets.

66 spacecraft the entire operational Iridium constellation identified via physical RF signatures
5.17 million Ring Alert messages collected continuously over 32 consecutive days
97.75% average spacecraft identification accuracy over 25 unseen days after training
0.9715 AUROC ability to reject an unseen terrestrial SDR transmitter spoofing satellite IDs

The Signal That Cannot Lie

When a satellite in Low Earth Orbit (780 km altitude) broadcasts an alert to the ground, the packet contains a simple numeric identifier: „I am satellite 42”. For an adversary on the ground equipped with a commodity Software-Defined Radio, it is trivial to construct an identical bitstream claiming to be satellite 42.

The distinction resides in the analog RF chain. No two power transistors are atomically identical, no crystal oscillator operates without its own subtle phase jitter, and microscopic gain imbalances between the In-Phase (I) and Quadrature (Q) branches deform the modulation constellation in a repeatable manner. These analog variations form the transmitter's Specific Emitter Identification (SEI). Whatever the digital packet claims, the receiver can analyze the waveform geometry to verify whether the hardware matches the legitimate satellite.

Vector Signal Analyzer: I/Q Constellation & Hardware Fingerprints
L-Band · 1616.0 – 1626.5 MHz
Iridium-42 Orbital Plane 1
I/Q Origin Offset
I: +4.2% | Q: -2.8%
Phase Jitter Noise
1.45° RMS
Amplifier Compression
0.962
Received Orbit
780 km LEO
Power amplifier exhibits slight I-channel asymmetry (+4.2%) and 0.962 non-linear compression at peak burst.

The Spoofer Test: When Bits Lie but Physics Refuses

To evaluate whether this approach can protect operational systems, researchers conducted an over-the-air attack experiment. A ground operator utilized a commodity Software-Defined Radio to broadcast Ring Alert frames formatted in full compliance with the Iridium protocol standard, embedding the satellite identifier of Iridium-42.

The neural classifier had been trained exclusively on genuine satellite bursts received from space; the ground SDR was entirely absent from training data. In this open-set test, the detector achieved an AUROC of 0.9715 in separating genuine spacecraft from the terrestrial impersonator. While the decoded packet header appeared authentic, the physical waveform geometry instantly betrayed the terrestrial origin of the signal.

Protocol Layer (What the Packet Claims)
PACKET_HEADER: 0x5A4F
CLAIMED_SATELLITE_ID: IRIDIUM-42
BURST_TYPE: RING_ALERT_L_BAND
CRC_CHECKSUM: VALID (PASS)
STATUS: PROTOCOL_COMPLIANT
All digital fields are valid and fully conform to communication specifications.
RF Verdict: Genuine Spacecraft
ANALOG_PROFILE: GENUINE_SV-042_MATCH
IQ_DISTORTION_METRIC: NOMINAL_SPACE
CLASSIFIER_OUTPUT: Identification Accuracy: 97.75%
DECISION: HARDWARE SIGNATURE CONFIRMED
Waveform geometry directly matched against the constellation hardware profile library.

The AI Shortcut Trap: Learning the Radio vs. Learning the Room

The most consequential finding of the benchmark is not classifier accuracy, but a critical warning regarding machine learning applied to physical signals.

When researchers trained a classifier using transmissions recorded exclusively inside an enclosed office, the model scored a near-perfect 0.991 AUROC within that same room. However, when the spoofing transmitter was moved outdoors to a 50-meter distance behind terrain obstructions, performance collapsed to 0.458 AUROC — worse than random chance.

The model had not learned the hardware signature of the transmitter; it had memorized the wall reflections, room resonances, and multipath echoes of that specific room. The purported spoof detector had inadvertently become a „room detector”.

Environmental Generalization & Channel Robustness Matrix
The model memorized room reflections and wall multipath. Outdoors, the 0.458 AUROC is worse than a coin flip.

The remedy required deliberate propagation diversity: training the classifier across indoor, hallway, and outdoor propagation channels forced the neural network to discard environmental echoes and focus exclusively on invariant transmitter hardware features, maintaining robust performance above 0.965 AUROC across all environments.

Strategic Relevance and Operational Boundaries

Iridium is not an experimental network; it is an operational backbone for global military and governmental infrastructure. Under its $738.5 million Enhanced Mobile Satellite Services (EMSS) contract with the U.S. Space Force, the constellation provides unconstrained airtime for tactical communications, command-and-control, and situational awareness across allied forces (including at RIMPAC 2026, where tracking units provided a common operational picture across vessels from 30 partner nations).

Authenticating the physical hardware origin of a transmission beyond its message contents provides an essential layer of electronic protection (Specific Emitter Identification). However, the boundary is precise: this research does not expose a vulnerability in military encryption or compromise the dedicated secure gateway. It proves that publicly receivable RF signaling waveforms can be authenticated through their underlying physics, offering an independent sensor when digital packet claims cannot be inherently trusted.

Methodological Note

Empirical datasets and experimental results are derived from the research paper IriSig-Spoof: A Real-World Benchmark for Time-Robust Satellite RF Fingerprinting and Spoofing Detection (arXiv:2608.18642v1, published 19 August 2026).

The primary dataset comprises 5,170,000 L-band Ring Alert transmissions (1616.0–1626.5 MHz) collected on a rooftop receiver at Xidian University in Xi’an between 27 June and 28 July 2025 (744 GB raw I/Q baseband data). Reference classifiers were trained on the first 7 days and evaluated on the subsequent 25 days to evaluate long-term temporal signature stability.