QUASAR: A Quantum-Classical Neural Network for SAR Satellite Physical-Layer Authentication. SAR stands for synthetic aperture radar. The authentication problem is whether hardware fingerprints can distinguish X-band satellite transmissions, even during replay, crafted-signal injection and space-borne spoofing. In this 2026 arXiv study, Vincenzo Sammartino, Nathanael Denis and Roberto Di Pietro test a classifier that combines classical and quantum components in that setting.

Physical-layer authentication checks whether a signal carries the expected transmitter-specific hardware imperfections. In this setting, those imperfections appear in radio samples as IQ imbalances, phase micro-perturbations, oscillator drift and mixer nonlinearity. This supplies evidence separate from cryptographic authentication, which may be unavailable or compromised. The challenge is especially sharp within one satellite constellation because units can share the same bus and payload design. QUASAR targets X-band signals, spanning 8–12 GHz.

The research question is whether QUASAR can learn an enrolled satellite’s fingerprint well enough to make a binary authentication decision: did this burst come from the target satellite, or not? The evaluation asks whether that decision survives recorded-burst replay, synthetic signal injection and spoofing by another satellite. This is useful because accepting an impersonator could admit fabricated radar data, while rejecting legitimate transmissions could disrupt observation services. Similar hardware across satellites makes intra-constellation discrimination particularly challenging.

The researchers built an X-band capture testbed using a programmable downconversion mixer and software-defined radios. They collected transmissions from 37 ICEYE satellites over 28 days and used independent receivers to examine transfer between acquisition devices. Enrollment used verified satellite passes. QUASAR then fused two components. A convolutional neural network encoded broad patterns from signal spectrograms. A variational quantum circuit received complex radio samples mapped into qubit states while preserving their magnitude and phase. Entangling operations coupled those states so the quantum branch could learn interactions across a block of samples.

QUASAR reached 97.3% validation accuracy, exceeding the classical-only baseline. It also matched classical baselines while using only a small fraction of the training data. For that reduced-data comparison, all baselines were trained on the same subset, making the data budget consistent across models. These results establish comparative performance within the evaluated benchmark; they do not, by themselves, show that the same advantage will persist under different receivers, constellations or collection periods.

The attack evaluation covered replay attacks, crafted-IQ injections and space-borne spoofing attempts. Detection was highest for crafted-IQ injections, followed by replay attacks, and lowest for space-borne spoofing. The space-borne test treated another ICEYE satellite as the attacker, creating a difficult comparison because units in the constellation share similar architecture and hardware. The measurements do not establish why performance varied across the attack types.

The evaluation used ICEYE satellites over 28 days, so transfer to other constellations and longer operating periods remains uncertain. Hardware fingerprints may drift and require periodic model updates. The analysis also warns that one trained fingerprint cannot simply substitute for another, leaving transfer across receivers or collection chains as an operational concern. Training the hybrid architecture is computationally intensive because its parameter-shift method repeatedly evaluates the quantum circuit for each trainable quantum parameter. That cost affects retraining even if authentication after enrollment can operate near real time.

Satellite ground-station defenders could evaluate QUASAR as a defense-in-depth signal, not as a replacement for cryptographic controls. A cautious pilot would enroll the system with verified passes captured through the intended receiver chain, then test replay, crafted injection and satellite-spoofing conditions locally. Teams should also monitor fingerprint drift and plan for retraining after receiver or acquisition-chain changes. They should not assume the reported detection rates or training-data advantage will transfer unchanged to their own constellation and equipment.

QUASAR contributes an implemented X-band fingerprinting testbed and a hybrid classifier that combined conventional spectral encoding with a variational quantum circuit. In the evaluated ICEYE setting, it improved validation accuracy, used less training data and detected all three tested attack classes at differing rates. Satellite security architects and detection engineers have reason to investigate the approach through local trials. They should retain existing controls and avoid interpreting this single-constellation campaign as proof of portable, long-term authentication performance.