Specific Emitter Identification (SEI)

The analysis of errors in the in-phase (I) and quadrature (Q) components of a digitally modulated signal. These errors, including I/Q gain imbalance, quadrature skew, and DC offset, are caused by imperfect analog mixers and local oscillators. Because these imperfections are unique to each transmitter's physical hardware, they form a highly discriminative and unclonable fingerprint for SEI systems.

A technique that exploits the periodic statistical properties of modulated signals. Unlike stationary noise, communication signals exhibit cyclostationarity tied to the symbol rate, carrier frequency, and pulse shaping. By computing the spectral correlation function (SCF), SEI systems can extract features that are robust to stationary interference and reveal subtle, device-specific variations in signal generation.

A self-supervised deep learning paradigm that trains a model to learn robust feature embeddings without explicit labels. The model is trained to pull representations of signal samples from the same device closer together in the latent space while pushing samples from different devices apart. This approach is ideal for SEI because it learns to ignore channel noise and focus on the intrinsic hardware impairments that define a device's identity.

A classification methodology essential for real-world SEI deployment. Unlike traditional closed-set classifiers that assume all test signals belong to a known device, open set recognition reliably detects and rejects unknown or rogue transmitters never seen during training. Algorithms like OpenMax replace the standard SoftMax layer with a calibrated rejection mechanism based on Extreme Value Theory (EVT) to model the probability of novelty.

A subfield of transfer learning that addresses the distribution shift between training and operational data. An SEI model trained in a lab environment will fail if deployed in a new channel with different multipath characteristics. Domain adaptation techniques, such as adversarial domain alignment, force the model to learn channel-invariant features that represent only the transmitter's hardware signature, ensuring robust performance across diverse environments.

A higher-order statistic that computes the Fourier transform of the third-order cumulant of a signal. The bispectrum is uniquely valuable for SEI because it suppresses all Gaussian noise while preserving the phase information lost in standard power spectrum analysis. This reveals non-linear harmonic coupling caused by amplifier non-linearities, providing a rich, noise-robust feature set for identifying specific emitters.