Cyclic Cumulant

The cyclic cumulant generating function (CCGF) is the logarithm of the moment generating function evaluated at cyclic frequencies α. It provides the theoretical link between time-varying moments and cumulants.

• Defined as the coefficient of the Taylor series expansion of the log-characteristic function
• Cyclic frequencies α correspond to integer multiples of the symbol rate, carrier frequency, or their combinations
• For a signal x(t), the n-th order cyclic cumulant at cycle frequency α is denoted C_{n,x}^α(τ₁,...,τ_{n-1})
• Key property: Gaussian noise has zero cyclic cumulants for n > 2, enabling noise-robust feature extraction
• The CCGF decomposes into a sum over all partitions of the index set, relating moments to cumulants through the Leonov-Shiryaev formula

Cyclic cumulants exhibit frequency-selective behavior in the cycle frequency domain, meaning they are non-zero only at specific α values determined by the signal's underlying periodicities.

• For linearly modulated signals (QAM, PSK, PAM), cyclic frequencies occur at α = k/T_sym where T_sym is the symbol period
• Different modulation schemes produce distinct cyclic frequency patterns in their higher-order cumulant functions
• The cyclic cumulant at order n and frequency α is estimated via synchronized averaging: Ĉ_{n,x}^α = (1/T) ∫ x(t)x(t+τ₁)...x(t+τ_{n-1}) e^{-j2παt} dt
• Cycle frequency resolution is inversely proportional to observation time, requiring sufficient data for reliable estimation
• This selectivity enables blind modulation classification by matching observed cyclic frequency patterns to known modulation templates

Cyclic cumulants possess the additivity property for statistically independent signal components, making them powerful for blind source separation in co-channel interference scenarios.

• For independent signals x(t) and y(t), the cyclic cumulant of their sum satisfies: C_{n,x+y}^α = C_{n,x}^α + C_{n,y}^α
• This holds only when x and y are mutually independent and at least one has zero mean
• Additivity enables cumulant-based interference rejection: desired signal cumulants can be isolated even when overlapping in time and frequency
• The property extends to convolutive mixtures when combined with cyclic Wiener filtering
• Practical application: separating multiple co-channel emitters in dense electromagnetic environments without prior spatial information

Unlike second-order cyclic statistics, higher-order cyclic cumulants preserve phase information and detect non-linear phase coupling between spectral components.

• Cyclic polyspectra (Fourier transforms of cyclic cumulants) reveal quadratic and cubic phase coupling at specific cycle frequencies
• Phase coupling arises from transmitter non-linearities such as amplifier compression and mixer intermodulation
• The cyclic bispectrum S_{3x}^α(f₁,f₂) is non-zero only when f₁+f₂ = α, creating a two-dimensional frequency support plane
• Phase randomization techniques used in some modulation schemes (e.g., OQPSK) produce characteristic nulls in cyclic cumulant surfaces
• Hardware-specific phase coupling patterns serve as unique device fingerprints that persist across different transmitted data sequences

Sample cyclic cumulant estimates are asymptotically Gaussian distributed, enabling rigorous statistical hypothesis testing for modulation and device classification.

• The estimator Ĉ_{n,x}^α(τ) converges in distribution to a complex Gaussian as observation length T → ∞
• Asymptotic variance depends on both the cyclic cumulant order n and the cycle frequency α
• Covariance between estimates at different cycle frequencies is zero for distinct α values, simplifying multi-hypothesis testing
• Cyclic cumulant-based detection uses generalized likelihood ratio tests comparing estimated cumulant magnitudes against noise-only thresholds
• The asymptotic normality enables confidence interval construction for emitter identification decisions, critical for security applications requiring controlled false-alarm rates

Cyclic cumulants exhibit well-defined behavior under signal transformations, enabling normalized feature extraction robust to unknown channel gain and carrier frequency offset.

• Scale invariance: For a scaled signal y(t) = a·x(t), the n-th order cyclic cumulant scales as C_{n,y}^α = a^n · C_{n,x}^α
• Frequency shift: A carrier offset Δf shifts all cycle frequencies by n·Δf, preserving the relative cyclic frequency structure
• Time shift: A delay t₀ introduces a phase rotation e^{-j2παt₀} but preserves magnitude, enabling delay-invariant fingerprinting
• Normalization by appropriate cumulant powers yields scale-invariant features: C_{n,x}^α / (C_{2,x}^0)^{n/2}
• These invariance properties allow cyclic cumulant fingerprints to remain stable across varying receiver gain settings and coarse synchronization errors

Cyclic cumulants serve as discriminative features for automatic modulation classification without prior knowledge of carrier frequency or symbol rate. The cyclic cumulant magnitude at specific cycle frequencies forms a unique signature for each modulation scheme.

• QAM vs. PSK: Fourth-order cyclic cumulants at the symbol rate distinguish quadrature amplitude modulation from phase-shift keying
• Hierarchical classification: Higher-order cyclic cumulants enable differentiation between modulation families (e.g., 16-QAM vs. 64-QAM)
• Robustness: Features remain stable under multipath fading and carrier offset conditions that degrade conventional cumulant-based classifiers

Transmitter-specific non-linear distortions from power amplifiers, mixers, and DACs generate unique cyclic cumulant signatures that persist across different transmitted data sequences.

• Amplifier non-linearity: Third-order cyclic cumulants capture odd-order harmonic distortion patterns unique to each PA
• I/Q imbalance: Conjugate cyclic cumulants reveal the asymmetric gain and phase errors in quadrature modulators
• Phase noise: Cyclic cumulant dispersion at specific cycle frequencies quantifies oscillator instability for device discrimination
• DAC clock jitter: Periodic timing errors manifest as distinctive cyclic cumulant patterns at multiples of the sampling frequency

When multiple transmitters share the same frequency band, joint cyclic cumulant diagonalization exploits differences in their cyclostationary signatures to separate and identify individual emitters.

• Cycle frequency diversity: Each emitter's unique symbol rate and carrier offset creates separable cyclic cumulant subspaces
• Higher-order independence: Fourth-order cyclic cumulants provide sufficient statistics for blind source separation even when second-order methods fail
• Underdetermined mixtures: Cyclic cumulant tensors enable separation of more sources than sensors by exploiting non-Gaussianity and cyclostationarity jointly

Cyclic cumulant analysis detects signal retransmission artifacts introduced by relay attacks and spoofing devices that conventional power or correlation-based methods miss.

• Amplifier cascade signatures: A relay device's own PA introduces secondary non-linearities detectable through higher-order cyclic cumulants
• Phase discontinuity detection: Abrupt changes in cyclic cumulant phase reveal switching between genuine and spoofed transmissions
• Transient analysis: The turn-on transient of a spoofing device generates a burst of non-Gaussian cyclostationary energy distinguishable from the legitimate emitter's steady-state signature

Cyclic cumulants exhibit inherent immunity to stationary Gaussian noise and can be made robust to multipath through appropriate cycle frequency selection and normalization.

• Gaussian noise suppression: All cyclic cumulants of order > 2 are theoretically zero for Gaussian processes, providing natural noise rejection
• Multipath invariance: Cyclic cumulants at cycle frequencies that are integer multiples of the symbol rate remain proportional to the original values under frequency-selective fading
• Bicoherence normalization: Dividing the cyclic bispectrum by the power spectrum at each cycle frequency yields a scale-invariant feature robust to varying received signal strength

Cyclic cumulant-based one-class classifiers enable detection of previously unseen transmitters by modeling the statistical boundaries of known device signatures in higher-order feature space.

• Novelty detection: Extreme value theory applied to cyclic cumulant distributions defines thresholds for rejecting unknown emitters
• Incremental enrollment: New device signatures are added by updating cyclic cumulant templates without retraining the entire classifier
• Drift tracking: Slow temporal variations in cyclic cumulant features due to temperature and aging are modeled as trajectories in cumulant space, enabling continuous authentication

A third-order frequency-domain representation that detects quadratic phase coupling between signal components. Unlike the power spectrum, the bispectrum reveals non-Gaussian signatures and phase relationships that are invisible to second-order statistics. It is computed as the 2D Fourier transform of the third-order cumulant sequence, producing a bifrequency plane where peaks indicate non-linear interactions between frequency pairs. This makes it a foundational tool for identifying transmitter-specific harmonic generation and intermodulation products.

Statistical measures beyond second-order variance that quantify deviations from Gaussianity in signal amplitude distributions. Key cumulants include:

• Third-order (skewness): captures distributional asymmetry caused by amplifier non-linearity
• Fourth-order (kurtosis): measures tailedness, sensitive to impulsive hardware distortions
• Theoretical property: Gaussian processes have zero cumulants of order >2, enabling robust noise suppression These form the mathematical backbone for constructing feature vectors that are inherently resilient to additive white Gaussian noise.

A normalized bispectrum that measures the proportion of signal energy at a bifrequency pair that is quadratically phase-coupled. Bounded between 0 and 1, bicoherence provides a consistent metric for non-linearity detection regardless of signal power. Values near 1 indicate strong deterministic phase coupling characteristic of specific transmitter impairments, while values near 0 suggest random phase relationships. This normalization makes bicoherence particularly valuable for comparing signatures across devices operating at different power levels.

A compact statistical fingerprint constructed from estimated higher-order cumulants that serves as input to machine learning classifiers. Typical construction involves:

• Estimating cumulants up to 4th order from signal segments
• Organizing estimates into a fixed-dimensional vector
• Applying higher-order whitening to decorrelate features
• Using the resulting vector with classifiers such as SVMs or neural networks These vectors capture hardware-specific non-linearities while suppressing Gaussian noise, enabling robust device discrimination even at low signal-to-noise ratios.

A non-linear phenomenon where two frequency components f1 and f2 interact within a device's analog circuitry to generate a third component at their sum or difference frequency (f1±f2). This coupling arises from amplifier non-linearities, mixer imperfections, and other hardware impairments. The resulting phase-coherent harmonic serves as a distinctive, unclonable signature because the exact coupling strength and phase relationship depend on the unique physical properties of each transmitter's components.

The exploitation of higher-order statistics' theoretical insensitivity to Gaussian processes to extract non-Gaussian signal features buried below the noise floor. Since Gaussian distributions have zero cumulants of order greater than two, cumulant-based processing mathematically eliminates additive white Gaussian noise. This property enables:

• Feature extraction at negative SNR conditions
• Robust operation in contested electromagnetic environments
• Reliable identification without requiring high-gain antennas or proximity This is a key advantage over traditional power-spectrum-based methods.