Cyclostationary Analysis

The Spectral Correlation Density function is the fundamental tool of cyclostationary analysis, representing the correlation between spectral components of a signal separated by a specific cyclic frequency.

• Reveals hidden periodicities not visible in standard power spectral density
• Peaks in the SCD occur at cycle frequencies (α) corresponding to the signal's symbol rate, carrier frequency, and their harmonics
• Provides a two-dimensional signature (frequency f vs. cycle frequency α) unique to each modulation type
• Enables discrimination between overlapping signals that share the same frequency band but have different cyclic features

A defining advantage of cyclostationary analysis is its inherent immunity to stationary noise and interference.

• Stationary Gaussian noise exhibits no cyclic correlation at non-zero cycle frequencies
• The SCD function naturally separates modulated signals from background noise in the cyclic frequency domain
• Enables reliable feature extraction at negative SNR conditions where energy detection fails
• Interference signals with different symbol rates or carrier frequencies appear at distinct cyclic frequencies, allowing clean separation

Cyclostationary analysis enables the estimation of critical signal parameters without prior knowledge or demodulation.

• Symbol rate estimation: The cyclic frequency at which the strongest SCD peak appears corresponds directly to the baud rate
• Carrier frequency offset: Phase shifts in the cyclic autocorrelation reveal the offset between transmitter and receiver oscillators
• Timing recovery: The cyclic autocorrelation magnitude peaks at optimal sampling instants
• Modulation identification: The unique SCD pattern serves as a fingerprint for automatic modulation classification

The Cyclic Autocorrelation Function is the time-domain counterpart of the SCD, measuring the correlation between a signal and a frequency-shifted version of itself.

• Computed as the Fourier coefficient of the time-varying autocorrelation function
• Non-zero values at specific cycle frequencies indicate the presence of cyclostationarity
• Forms a Fourier transform pair with the Spectral Correlation Density function
• Often used as a computationally lighter alternative to full SCD estimation for real-time applications

Cyclostationary signatures provide highly discriminative features for Automatic Modulation Classification systems.

• BPSK signals exhibit strong cyclic features at twice the carrier frequency and at the symbol rate
• QAM signals show distinct SCD patterns based on constellation symmetry and order
• Higher-order cyclic cumulants are theoretically immune to Gaussian noise
• Feature vectors extracted from the SCD can be fed into deep neural networks for robust, blind classification
• Outperforms energy-based and raw I/Q classifiers in low-SNR and interference-limited scenarios

The Spectral Coherence is a normalized version of the SCD that quantifies the degree of cyclostationarity on a scale from 0 to 1.

• Normalizes the SCD by the geometric mean of the power spectral density at the two correlated frequencies
• Provides a magnitude-independent measure of cyclic correlation strength
• Useful for setting detection thresholds that adapt to varying signal power levels
• Enables consistent feature extraction across dynamic range variations common in real-world spectrum monitoring

The spectral correlation density is the core two-dimensional transform quantifying cyclostationarity. It measures the spectral correlation between a signal's frequency components separated by α/2, where α is the cycle frequency. Unlike the power spectral density, the SCD reveals hidden periodicities by showing how energy at one frequency correlates with energy at another. Key properties:

• Exhibits discrete peaks at cycle frequencies corresponding to the signal's symbol rate, carrier offset, and coding schemes
• Provides a noise-immune signature because stationary noise has no spectral correlation (SCD is zero for α ≠ 0)
• Computed efficiently using the FAM (FFT Accumulation Method) or SSCA (Strip Spectral Correlation Analyzer) algorithms
• Used as a robust feature map input to convolutional neural networks for automatic modulation classification

The cycle frequency (α) is the fundamental parameter defining the periodicity of a signal's second-order statistics. It represents the spacing between correlated spectral components. For a signal with symbol rate Rs and carrier frequency fc:

• Symbol rate harmonics: α = k·Rs (k = 0, ±1, ±2, ...)
• Carrier-related: α = ±2fc + k·Rs
• Coding and framing: α related to frame rates, spreading codes, or pilot patterns Identifying the set of cycle frequencies is the primary goal of cyclic feature detection, enabling blind estimation of signal parameters without demodulation. The presence of a cycle frequency at α ≠ 0 definitively distinguishes a modulated signal from stationary noise.

The cyclic autocorrelation function is the time-domain dual of the spectral correlation density. It measures the correlation between a signal and a frequency-shifted version of itself over time. Formally defined as:

• R_x^α(τ) = E{x(t + τ/2)x*(t - τ/2)e^{-j2παt}}
• For α = 0, it reduces to the conventional autocorrelation function
• Peaks in the CAF at specific delays τ and cycle frequencies α reveal the signal's baud rate, pulse shape, and timing offset
• The Fourier transform of the CAF with respect to τ yields the spectral correlation function, establishing the Wiener-Khintchine-like relationship for cyclostationary processes

Cyclic cumulants extend cyclostationary analysis beyond second-order statistics to higher-order moments (third, fourth, and beyond). They exploit the fact that modulated signals exhibit cyclostationarity in their higher-order probability distributions. Advantages over second-order methods:

• Immunity to Gaussian noise of any color, as Gaussian processes have zero cumulants of order > 2
• Ability to classify modulation formats that share the same second-order cycle frequencies (e.g., QPSK vs. 16-QAM)
• The fourth-order cyclic cumulant at specific cycle frequencies provides a unique fingerprint for each linear digital modulation scheme
• Robust to phase and frequency offsets when using normalized cumulant ratios

In Orthogonal Frequency Division Multiplexing (OFDM) systems, the cyclic prefix (CP) introduces a specific form of cyclostationarity. Because the CP is a copy of the end of the OFDM symbol prepended to the beginning, the signal exhibits correlation at a delay equal to the useful symbol length Tu. This manifests as:

• A cyclic autocorrelation peak at τ = Tu and α = k/Ts, where Ts = Tu + Tcp is the total symbol duration
• Enables blind OFDM parameter estimation: useful symbol duration, CP length, and symbol timing
• Allows distinguishing OFDM from single-carrier modulations without demodulation
• The CP-induced cyclostationarity is intentionally exploited in spectrum sensing for cognitive radio to detect primary users like LTE and Wi-Fi

The FFT Accumulation Method is the most widely used computationally efficient algorithm for estimating the spectral correlation density. It transforms the complex time-smoothing operation into a sequence of FFTs:

• Channelization: Input data is segmented and processed through a bank of narrowband filters via FFT
• Decimation: Each channel is downsampled to reduce computational load
• Cross-correlation: Product sequences between frequency-shifted channels are computed and averaged over time
• The FAM trades off cycle frequency resolution against spectral frequency resolution through the channelizer design
• Implemented in real-time SDR frameworks like GNU Radio for practical cyclostationary feature extraction