Spectral Coherence Function

The Spectral Coherence Function (SCF) is defined as the Spectral Correlation Function normalized by the geometric mean of the signal's power at two frequency-shifted components. This normalization constrains its magnitude to the interval [0, 1], where:

• 1 indicates perfect correlation between frequency-shifted components
• 0 indicates complete absence of correlation This bounded output makes it a unitless metric that is independent of absolute signal power, enabling direct comparison across different signal-to-noise ratio (SNR) conditions.

The SCF is formally expressed as:

C_x^α(f) = S_x^α(f) / √[S_x^0(f + α/2) · S_x^0(f - α/2)]

Where:

• S_x^α(f) is the Spectral Correlation Function at cyclic frequency α and spectral frequency f
• S_x^0(f ± α/2) are the stationary power spectral densities at the shifted frequencies
• The denominator acts as a normalization factor that removes the influence of the signal's power spectrum This formulation reveals the pure degree of correlation between spectral components separated by the cyclic frequency α.

A key advantage of the SCF over the unnormalized Spectral Correlation Function is its inherent resilience to colored noise and flat-fading channels. Because stationary noise exhibits no cyclostationarity, its contribution to the numerator approaches zero while the denominator accounts for its power:

• Stationary Gaussian noise produces an SCF value near zero at all non-zero cyclic frequencies
• Narrowband interference appears as a distinct, localized peak in the SCF plane
• Multipath fading scales both numerator and denominator, preserving the coherence magnitude This property makes the SCF a preferred feature for blind modulation classification in contested electromagnetic environments.

The SCF is typically visualized as a two-dimensional bifrequency plane with axes:

• Spectral frequency (f): The conventional frequency axis
• Cycle frequency (α): The separation between correlated components

Key visual features include:

• Alpha profile: A horizontal slice at a fixed f, revealing the strength of cyclostationarity across all α
• Spectral frequency profile: A vertical slice at a fixed α, showing which frequency bands exhibit correlation
• Coherence peaks: Distinct local maxima at specific (f, α) pairs that form a unique cyclic signature for each modulation scheme This representation allows engineers to visually identify modulation-specific patterns even in low-SNR conditions.

The SCF serves as a discriminative feature space for automatic modulation classification systems. Different digital modulation schemes produce distinct patterns of coherence peaks:

• BPSK exhibits strong coherence at α = 2f_c (twice the carrier frequency) and α = symbol rate
• QPSK shows peaks at α = symbol rate but suppressed features at 2f_c due to phase randomization
• MSK/GMSK generates unique coherence patterns at α = half the symbol rate
• OFDM produces a characteristic ridge at α = subcarrier spacing due to cyclic prefix repetition These patterns are extracted as cyclic feature vectors and fed into neural network classifiers for robust, blind identification.

Practical computation of the SCF relies on the FFT Accumulation Method (FAM), a computationally efficient channelizer-based approach:

1. The input signal is decomposed into frequency channels via a short-time FFT
2. Channel outputs are frequency-shifted and complex-conjugated according to the desired α
3. The product of shifted channels is time-averaged to estimate the spectral correlation
4. The result is normalized by the power estimates from the zero-cycle-frequency plane

The FAM algorithm reduces computational complexity from O(N²) to O(N log N), making real-time SCF estimation feasible on FPGA and GPU hardware for deployed spectrum monitoring systems.

The foundational two-dimensional transform from which the Spectral Coherence Function is derived. The SCF measures the raw correlation between frequency-shifted versions of a signal.

• Represents correlation density as a function of spectral frequency (f) and cyclic frequency (α)
• The Spectral Coherence Function normalizes the SCF by the signal's power, yielding a value between 0 and 1
• Essential for distinguishing between second-order cyclostationary signals and stationary noise

The key parameter, denoted by α, that defines the periodicity of a signal's statistical moments. The Spectral Coherence Function is non-zero only at specific α values.

• For a signal with symbol rate Rs, strong coherence peaks appear at α = k * Rs (where k is an integer)
• Carrier frequency offsets manifest as shifts in the α domain
• The pattern of non-zero α values forms a unique cyclic signature for each modulation type

A scalar metric that quantifies the relative strength of a signal's cyclostationary features compared to its total power. Directly derived from the Spectral Coherence Function.

• Ranges from 0 (completely stationary) to 1 (perfectly cyclostationary)
• Used as a threshold to determine if a signal of interest is present in a noisy channel
• Provides a single-number summary for blind modulation classification without requiring full cyclic spectrum analysis

A compact set of features extracted from the Spectral Coherence Function at specific cycle frequencies (α) and spectral frequencies (f).

• Reduces the high-dimensional SCF/coherence data to a manageable input for neural network classifiers
• Typically includes the magnitude of coherence peaks at α = Rs, 2Rs, and 4Rs
• Robust to channel impairments like fading and noise when normalized coherence values are used

A statistical hypothesis test formulated in the frequency domain to detect the presence of cyclostationarity at a specific cyclic frequency (α).

• Uses the Spectral Coherence Function to determine if an observed peak is statistically significant or a random fluctuation
• Provides a constant false alarm rate (CFAR) detection framework
• Forms the theoretical basis for many blind parameter extraction algorithms that rely on coherence thresholding

A Frequency-Shift (FRESH) filtering technique that exploits the cyclostationarity measured by the Spectral Coherence Function to separate overlapping signals.

• Processes multiple frequency-shifted versions of the input, weighted by their coherence
• Can extract a weak signal of interest even when it shares the same frequency band as a stronger interferer
• The optimal FRESH filter weights are derived directly from the cyclic spectrum and coherence function