Cyclostationary Analysis

The SCF is the fundamental tool of cyclostationary analysis, representing the density of temporal correlation between spectral components separated by a specific cycle frequency (α).

• Mechanism: Computed as the Fourier transform of the cyclic autocorrelation function.
• Key Insight: While stationary noise has spectral correlation only at α=0, modulated signals exhibit non-zero correlation at cycle frequencies corresponding to their symbol rate, carrier frequency, and guard intervals.
• Example: A BPSK signal with a 1 MHz symbol rate will show spectral correlation peaks at α = 2fc (twice the carrier) and α = 1 MHz (the symbol rate).

Cycle frequencies (α) are the specific periodicities at which a signal's statistical properties repeat. This domain provides a signature fingerprint for each modulation scheme.

• Separation: Signals overlapping in both time and frequency can be cleanly separated in the cycle frequency domain because each modulation type generates a unique set of α values.
• Robustness: These features are deterministic and persist even when the signal is buried below the noise floor, making the technique exceptionally robust for low-SNR environments.
• Common Cycle Frequencies: Symbol rates, chip rates, frame rates, and twice the carrier frequency.

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

• Theoretical Basis: Wide-sense stationary (WSS) noise exhibits no spectral correlation for α ≠ 0. By analyzing the SCF at non-zero cycle frequencies, the noise contribution is mathematically eliminated.
• Practical Impact: This allows for signal detection and parameter estimation at SNRs where a conventional energy detector or PSD analyzer would fail completely.
• Interference Discrimination: Co-channel interferers with different symbol rates or modulation types are easily distinguished by their distinct cycle frequencies.

Cyclostationary analysis enables the extraction of critical signal parameters without any prior knowledge of the transmission.

• Carrier Frequency: Estimated from the location of SCF peaks at α = 2fc.
• Symbol/Keying Rate: Directly measured from the cycle frequency of the dominant spectral correlation feature.
• Timing Recovery: The phase of the cyclic autocorrelation provides information for synchronizing to the symbol clock.
• Application: This is critical for automatic modulation classification (AMC) and cognitive radio systems that must adapt to unknown emitters.

In spectrum monitoring, an established cyclostationary profile serves as a highly sensitive baseline for anomaly detection.

• Baseline Construction: A normal signal's SCF or cyclic domain profile (CDP) is characterized during a training phase.
• Anomaly Trigger: Any deviation—such as a new cycle frequency appearing, a shift in existing α values, or a change in correlation strength—indicates a potential anomaly.
• Use Case: Detecting a rogue emitter that has hijacked a legitimate frequency or identifying a malfunctioning transmitter whose timing has drifted, all without demodulating the signal.

The FAM is the most widely used efficient algorithm for computing the SCF, making real-time cyclostationary analysis feasible.

• Process: It uses a channelizer to slice the spectrum, followed by decimation and a series of complex FFT operations to compute cross-spectral correlation.
• Trade-off: FAM provides a balance between computational load and the resolution of the SCF in both the spectral frequency (f) and cycle frequency (α) domains.
• Alternative: The Strip Spectral Correlation Analyzer (SSCA) is another algorithm optimized for different resource constraints, often used in FPGA implementations.

The Spectral Correlation Function is the fundamental mathematical tool of cyclostationary analysis. It measures the spectral correlation between a signal's frequency components separated by a specific cyclic frequency (α). Unlike the standard Power Spectral Density (PSD), which is a one-dimensional function of frequency, the SCF is a two-dimensional function of both frequency (f) and cyclic frequency (α). A signal exhibits cyclostationarity if the SCF is non-zero for some α ≠ 0. This property allows the separation of overlapping signals that share the same PSD but have different cyclic features, such as distinct symbol rates or carrier frequencies.

The Cyclic Autocorrelation Function is the time-domain counterpart of the SCF. It measures the correlation between a signal and a frequency-shifted version of itself over time. For a cyclostationary signal, the autocorrelation function is periodic in time, meaning R_x(t, τ) = R_x(t + T₀, τ) for some fundamental period T₀. This periodicity is directly tied to the signal's underlying physical parameters:

• Symbol rate for digitally modulated signals
• Carrier frequency offsets
• Chip rate for spread-spectrum signals
• Frame or slot rates in TDMA systems

A cyclic frequency is the specific frequency at which a signal's statistical properties exhibit periodicity. For a modulated signal, cyclic frequencies are typically integer multiples of the symbol rate, carrier frequency offset, or combinations thereof. The set of cyclic frequencies for a given signal forms a unique signature that can be used for:

• Signal identification: Each modulation type (BPSK, QPSK, 16-QAM) has a distinct cyclic frequency pattern.
• Parameter estimation: The symbol rate can be directly estimated from the cyclic frequency.
• Interference separation: Signals with different symbol rates occupy distinct cyclic frequency planes, enabling their separation even when they overlap in both time and frequency.

Understanding the distinction between these two statistical states is critical:

• Stationary Process: A signal whose statistical properties (mean, autocorrelation) are time-invariant. White noise is a classic example. Its PSD fully characterizes it.
• Cyclostationary Process: A signal whose statistical properties vary periodically with time. Most man-made communication signals fall into this category due to modulation, coding, and framing.
• Non-Stationary Process: A signal with no periodic or time-invariant statistical structure.

Cyclostationary analysis exploits the fact that noise and interference are often stationary, while the signal of interest is cyclostationary, providing a powerful signal-selective processing gain.

While second-order cyclostationarity (exploiting the autocorrelation) is common, higher-order cyclic cumulants exploit the periodicity in third-order, fourth-order, and higher statistical moments. These are essential when:

• Second-order cyclic features are absent or weak.
• Signals share the same symbol rate but have different modulation orders (e.g., QPSK vs. 16-QAM).
• Gaussian noise must be completely suppressed, as Gaussian processes have zero cumulants of order greater than two.

Fourth-order cyclic cumulants are particularly powerful for classifying modulation types in low-SNR environments.

Cyclic Wiener filtering is an optimal linear filtering technique that exploits cyclostationarity for signal extraction. Unlike a standard Wiener filter, which is time-invariant, a cyclic Wiener filter is periodically time-varying. It can extract a desired cyclostationary signal from a mixture of stationary noise and other cyclostationary interferers by leveraging their distinct cyclic frequencies. This is also known as FRESH (FREquency-SHift) filtering, which implements the filter as a bank of frequency-shifted linear time-invariant filters, providing a practical architecture for real-time interference mitigation.