Cyclic Cumulant

A cyclic cumulant is a Fourier coefficient of a time-varying cumulant function. For a cyclostationary signal, the time-varying cumulant is a periodic function, and its Fourier series expansion reveals distinct spectral lines at cycle frequencies (α) where signal energy concentrates. The cyclic cumulant of order n and conjugation configuration (*) at cycle frequency α is defined as the Fourier coefficient of the time-varying cumulant function. This decomposition separates the signal's statistical structure from stationary Gaussian noise, which has no cyclic cumulant components at α ≠ 0.

The defining advantage of cyclic cumulants is their inherent immunity to stationary Gaussian noise. Because stationary noise has no periodicity in its statistics, its cyclic cumulants vanish for all non-zero cycle frequencies. This property enables modulation classification even at negative signal-to-noise ratios (SNR) where raw IQ samples or conventional cumulants fail. The cyclic cumulant estimator effectively filters out wide-sense stationary interference, making it indispensable for spectrum surveillance and cognitive radio applications in congested electromagnetic environments.

Each modulation scheme generates a unique cyclic cumulant signature in the cycle frequency domain. Key cycle frequencies are integer multiples of the symbol rate (k/T_sym) and combinations with the carrier frequency offset. For example:

• BPSK: Strong cyclic cumulant peaks at α = 2f_c and α = 2f_c ± 1/T_sym
• QPSK: Fourth-order cyclic cumulant peaks at α = 4f_c and α = 4f_c ± k/T_sym
• 16-QAM: Distinct fourth-order cyclic cumulant pattern at α = 4f_c This spectral structure enables blind modulation identification by matching observed cyclic cumulant peaks against theoretical templates.

Practical cyclic cumulant estimation uses the cyclic temporal cumulant (CTC) or cyclic polyspectrum (CPS) methods. The CTC approach computes time-varying cumulants over a sliding window and applies a Fourier transform to extract cyclic components. The CPS method estimates the cyclic cumulant directly in the frequency domain via the cyclic cumulant spectrum. Key implementation considerations:

• Observation length: Longer records improve cycle frequency resolution but increase computational load
• FFT-based algorithms: Enable efficient estimation using the strip spectral correlation analyzer
• Recursive estimators: Support streaming applications with online updates

Cyclic cumulants enable blind source separation and classification of overlapping signals in the same frequency band. Because different signals typically have distinct symbol rates and carrier frequencies, their cyclic cumulant signatures occupy different cycle frequency locations. This allows:

• Signal-selective classification: Extract and identify individual signals from a mixture by tuning to their unique cycle frequencies
• Interference rejection: Suppress co-channel interferers by selecting cycle frequencies where only the desired signal has energy
• Source enumeration: Count the number of active signals by detecting distinct cyclic cumulant peaks

Cyclic cumulants generalize conventional higher-order cumulants by adding the cycle frequency dimension. A conventional cumulant is the cyclic cumulant evaluated at α = 0. This relationship creates a hierarchical classification framework:

• Conventional cumulants (α = 0): Provide coarse modulation separation (PSK vs. QAM)
• Cyclic cumulants (α ≠ 0): Enable fine-grained discrimination within modulation families and provide additional robustness
• Cumulant invariants: Cyclic cumulant ratios that are independent of amplitude, phase, and timing offsets can be constructed for non-cooperative classification