Cyclostationary Feature Extraction

The foundational mechanism of cyclostationary analysis. Unlike stationary noise, modulated signals exhibit a periodic autocorrelation function where statistical properties like mean and variance vary cyclically with time. This periodicity is directly linked to the symbol rate and carrier frequency of the transmission.

The frequency-domain representation of cyclostationarity, computed via the Fourier transform of the cyclic autocorrelation. The SCD reveals spectral correlation at cycle frequencies (α). A signal exhibits cyclostationarity if and only if there exists a non-zero α for which the SCD is non-zero. Key properties:

• α = 0: Represents the standard power spectral density.
• α ≠ 0: Reveals hidden periodicities, such as the symbol rate for a BPSK signal.

A defining advantage of cyclostationary feature extraction is its inherent immunity to stationary noise. Thermal noise is wide-sense stationary, meaning its cyclic autocorrelation is zero for all α ≠ 0. By searching for features at non-zero cycle frequencies, the analysis effectively filters out the noise floor, enabling detection and classification of signals buried well below the noise.

Different digital modulation schemes generate unique cyclostationary signatures. These signatures act as a fingerprint for automatic modulation classification:

• BPSK: Strong features at cycle frequencies of ±2fc + kRs, where fc is the carrier frequency and Rs is the symbol rate.
• QPSK/QAM: Features are typically concentrated at cycle frequencies equal to integer multiples of the symbol rate (kRs).
• OFDM: Exhibits cyclostationarity induced by the cyclic prefix, creating features at the OFDM symbol rate.

Cyclostationary analysis enables the non-cooperative extraction of critical signal parameters without prior knowledge of the transmitter. By identifying the cycle frequencies present in the SCD, a receiver can directly estimate the symbol rate, carrier frequency offset, and guard interval length. This is essential for spectrum monitoring and cognitive radio applications.

Practical estimation of cyclostationary features relies on time-smoothing or frequency-smoothing algorithms. The FFT Accumulation Method (FAM) is a computationally efficient approach that uses a channelizer to compute the complex demodulates before cross-correlation. The trade-off between cycle frequency resolution and spectral frequency resolution is governed by the total observation time and the channelizer bandwidth.

The foundational mathematical transform for cyclostationary analysis. It computes the correlation of a signal with a frequency-shifted version of itself, revealing hidden periodicities.

• Single-variable function of lag parameter τ and cycle frequency α
• Peaks at α ≠ 0 indicate the presence of cyclostationarity
• Directly decomposes into the Spectral Correlation Function via Fourier transform
• Used to isolate modulation-specific signatures buried below the noise floor

A two-dimensional frequency-frequency representation that exposes the correlation between spectral components separated by the cycle frequency. It is the frequency-domain counterpart of the cyclic autocorrelation.

• Resolves features in both spectral frequency (f) and cycle frequency (α)
• Stationary noise maps to the α = 0 plane only, enabling robust signal detection
• Distinct SCD patterns serve as fingerprints for modulation classification
• Computed efficiently using the FAM (FFT Accumulation Method) or SSCA (Strip Spectral Correlation Analyzer)

The fundamental parameter defining the periodicity of a statistical moment. For modulated signals, cycle frequencies are typically integer multiples of the symbol rate, carrier frequency offset, or chip rate.

• Symbol rate (Rs) and 2× carrier frequency are the most common cycle frequencies
• Different modulation schemes exhibit unique sets of cycle frequencies
• A signal is pure stationary if autocorrelation is non-zero only at α = 0
• Conjugate cycle frequencies appear for non-circular modulations like BPSK and GMSK

A computationally efficient algorithm for estimating the spectral correlation density by using a channelizer-based approach. It trades off resolution for speed, making real-time cyclostationary processing feasible.

• Decimates the signal into narrowband frequency channels via a sliding FFT
• Cross-correlates channel outputs to estimate Spectral Correlation Density
• Resolution controlled by channel spacing (Δf) and dwell time (Δt)
• Forms the backbone of many real-time signal classifiers in electronic warfare systems

A statistical property where a complex signal's distribution is not rotationally invariant, meaning it is correlated with its own complex conjugate. This property generates conjugate cyclostationary features.

• Proper signals (e.g., QPSK, QAM) have zero pseudo-autocorrelation
• Improper signals (e.g., BPSK, GMSK, AM) require widely linear processing
• Conjugate cycle frequencies reveal asymmetries in the constellation diagram
• Exploiting impropriety doubles the feature space for signal identification

Higher-order cyclic statistics that extend cyclostationary analysis beyond second-order moments. Cyclic cumulants are immune to Gaussian noise of any color, providing extreme robustness in low-SNR environments.

• Fourth-order cyclic cumulants classify QAM constellations by order (16-QAM vs. 64-QAM)
• Gaussian noise suppression is total—theoretically zero for orders > 2
• Computed from cyclic temporal moment functions with proper conjugation configurations
• Essential for blind modulation classification in electronic support measures (ESM)