Specific Emitter Identification (SEI)

SEI relies on unintentional modulation artifacts—microscopic signal distortions that are a byproduct of manufacturing variances, not designed features. These include I/Q imbalance, carrier frequency offset (CFO), and phase noise. Because these impairments are physically inherent to the analog front-end, they cannot be cloned by an adversary simply copying a device's digital identity or MAC address.

The identification process is entirely passive. The SEI system does not need to interrogate the target device or inject any probe signals. It operates simply by observing ambient transmissions. This makes it ideal for signals intelligence (SIGINT) and spectrum enforcement, where the target is non-cooperative and unaware of the identification process.

A robust SEI system does not rely on a single feature. It aggregates a high-dimensional feature vector extracted from multiple domains:

• Time domain: Transient turn-on/turn-off characteristics.
• Frequency domain: Carrier offset and spectral regrowth.
• Joint time-frequency: Wavelet-based transient analysis.
• Statistical domain: Higher-order statistics (bispectrum, kurtosis). This multi-domain approach ensures resilience against noise and countermeasures.

A critical challenge is separating the device fingerprint from the channel fingerprint. Multipath propagation and Doppler shift can distort the signal. Modern SEI systems use domain adaptation and contrastive learning to learn channel-robust representations, ensuring the extracted identity remains stable whether the emitter is in an open field or a dense urban canyon.

Hardware impairments are not perfectly static. Local oscillator aging, thermal drift, and voltage fluctuations cause a device's signature to slowly evolve over hours or months. Production-grade SEI systems implement drift compensation algorithms that continuously update the stored baseline fingerprint, preventing a gradual increase in the false rejection rate (FRR).

In real-world electromagnetic environments, the system will constantly encounter unknown emitters not present in the training database. SEI classifiers must therefore support open set recognition, correctly identifying known devices while flagging novel transmitters as 'unknown' rather than misclassifying them. This is typically achieved using embedding space analysis with a calibrated distance threshold.