Cyclic Autocorrelation Function

The CAF is a quadratic transformation that measures the correlation between a signal and a frequency-shifted, conjugated version of itself. For a signal x(t), it is defined as the limit of a time-average:

• Definition: R_x^α(τ) = lim(T→∞) 1/T ∫ x(t + τ/2) x*(t - τ/2) e^{-j2παt} dt
• Dual Variables: Parameterized by both the lag τ (time delay) and cyclic frequency α (frequency shift)
• Stationary Special Case: When α = 0, the CAF reduces to the standard autocorrelation function of a wide-sense stationary process
• Non-Zero α: Reveals hidden periodicities in the signal's second-order statistics that are invisible to conventional power spectral density analysis

For digitally modulated signals, the CAF exhibits non-zero values only at a discrete set of cyclic frequencies that are directly tied to the signal's physical parameters:

• Symbol Rate Harmonics: α = k/T_sym, where T_sym is the symbol period and k is an integer, revealing the baud rate
• Carrier Frequency Offsets: α = ±2f_c + k/T_sym, enabling blind carrier frequency estimation
• Pulse Shaping Dependence: The exact cyclic frequency pattern is a unique signature of the modulation format and pulse shaping filter
• BPSK Example: Exhibits strong features at α = 2f_c and α = 2f_c ± 1/T_sym, while QPSK suppresses the symbol-rate cyclic feature at α = 1/T_sym

The CAF provides inherent immunity to stationary noise because stationary processes have no cyclostationary content at non-zero cyclic frequencies:

• Stationary Noise Nulling: For any wide-sense stationary noise process n(t), R_n^α(τ) = 0 for all α ≠ 0
• Signal-on-Noise Isolation: The CAF of a noisy signal x(t) + n(t) at α ≠ 0 equals the CAF of the signal alone, completely rejecting the stationary noise component
• Interference Separation: Signals with different symbol rates or carrier frequencies occupy distinct, non-overlapping cyclic frequencies, enabling signal-selective processing
• Practical Advantage: This property makes cyclostationary feature detectors significantly more robust than energy detectors in low-SNR environments

The CAF exists in two complementary forms that capture different types of cyclostationarity:

• Non-Conjugate CAF: R_x^α(τ) = ⟨ x(t + τ/2) x*(t - τ/2) e^{-j2παt} ⟩, sensitive to spectral correlation between frequency components separated by α
• Conjugate CAF: R_{xx*}^α(τ) = ⟨ x(t + τ/2) x(t - τ/2) e^{-j2παt} ⟩, sensitive to conjugate spectral correlation
• Modulation Discrimination: Real-valued modulations like BPSK and PAM exhibit strong conjugate cyclostationarity, while complex-valued modulations like QPSK and QAM primarily exhibit non-conjugate features
• Joint Analysis: The presence or absence of features in both CAF variants provides a powerful discriminant for automatic modulation classification

The CAF and the Spectral Correlation Function (SCF) form a Fourier transform pair with respect to the lag variable τ:

• Transform Pair: S_x^α(f) = ∫ R_x^α(τ) e^{-j2πfτ} dτ, where S_x^α(f) is the cyclic spectrum
• Frequency-Domain Interpretation: The SCF at cyclic frequency α measures the correlation between spectral components at frequencies f + α/2 and f - α/2
• Computational Efficiency: While the CAF is defined in the time domain, practical estimation is often performed in the frequency domain using the FAM algorithm or SSCA algorithm
• Dual Representation: The CAF provides insight into temporal correlation structure, while the SCF reveals the frequency-selective nature of cyclostationarity

For purely cyclostationary signals, the CAF exhibits a Fourier series expansion in the global time variable t, revealing its periodic structure:

• Fourier Series Representation: R_x(t, τ) = Σ_α R_x^α(τ) e^{j2παt}, where R_x(t, τ) is the time-varying autocorrelation
• Cyclic Cumulant Connection: Higher-order cyclic statistics, such as the cyclic cumulant, generalize this periodic decomposition to third-order and fourth-order moments
• LPTV System Output: When a stationary signal passes through a Linear Periodically Time-Varying (LPTV) system, the output CAF is a sum of scaled and frequency-shifted versions of the input autocorrelation
• Modulation as LPTV: Digital modulation inherently acts as an LPTV transformation, imprinting cyclostationary features onto the transmitted waveform

A two-dimensional transform that measures the correlation between frequency-shifted versions of a signal. The SCF reveals hidden periodicities in spectral content that are invisible to standard power spectral density analysis.

• Maps correlation across both spectral frequency (f) and cyclic frequency (α)
• Non-zero values at α ≠ 0 indicate cyclostationarity
• Forms the basis for the FAM algorithm and SSCA algorithm estimators

The parameter that quantifies the periodicity of a signal's statistical moments. For digitally modulated signals, cyclic frequencies typically appear at:

• Symbol rate multiples: α = k/T_sym
• Carrier frequency offsets: α = ±2f_c + k/T_sym
• Combinations of the above

Detecting peaks at these α values enables blind parameter extraction without prior knowledge of the transmission scheme.

The FFT Accumulation Method is the most widely used computationally efficient estimator for the spectral correlation function. It works by:

• Channelizing the input signal into narrowband components
• Computing short-time FFTs across time segments
• Correlating frequency-shifted outputs and accumulating results

Offers significant complexity reduction over direct cyclic periodogram computation while maintaining reliable cyclic feature estimates.

A higher-order statistic extending cumulants to cyclostationary signals. Provides critical robustness against Gaussian noise in modulation classification.

• Second-order: Equivalent to cyclic autocorrelation
• Fourth-order: Discriminates between QAM constellations (16-QAM vs 64-QAM)
• Gaussian noise suppression: Cumulants of order > 2 are zero for Gaussian processes

Essential for classifying higher-order modulations in low-SNR environments.

A normalized version of the SCF that quantifies correlation strength on a scale from 0 to 1. This normalization removes the influence of signal power, making it a robust feature for:

• Comparing cyclostationary strength across different signal-to-noise ratios
• Setting detection thresholds in the Dandawate-Giannakis test
• Building cyclic feature vectors invariant to received power levels

A coherence value near 1 indicates strong cyclostationarity at that (f, α) pair.

The process of estimating modulation parameters without prior knowledge by analyzing cyclostationary features:

• Symbol rate estimation: Detect α peaks at multiples of 1/T_sym in the cyclic autocorrelation
• Carrier frequency offset: Identify α = ±2f_c patterns in the cyclic spectrum
• Cyclic prefix detection: Exploit OFDM-induced cyclostationarity for symbol timing

Enables fully autonomous signal intelligence and cognitive radio adaptation.