Cyclostationary OFDM Signature

The cyclic prefix (CP) creates a deterministic periodicity by copying the end of each OFDM symbol to its beginning. This repetition generates spectral correlation peaks at specific cycle frequencies (α = k/Ts, where Ts is the symbol duration). The correlation lag equals the useful symbol length Tu, producing a distinct ridge in the spectral correlation density (SCD) function. This feature is exploitable even when the signal power is well below the noise floor, as thermal noise lacks cyclostationarity.

Known pilot symbols inserted at predetermined subcarrier positions create a man-made cyclostationary signature. These pilots—such as the cell-specific reference signals (CRS) in LTE or demodulation reference signals (DMRS) in 5G NR—exhibit a regular time-frequency grid pattern. The resulting spectral correlation reveals:

• Subcarrier spacing from frequency-domain periodicity
• Symbol timing from time-domain repetition intervals
• Transmit antenna configuration from pilot density variations

The spectral correlation density (SCD) function Sxα(f) measures correlation between spectral components separated by α/2. For OFDM signals, distinct peaks emerge at:

• Cycle frequency α = k/Ts: CP-induced periodicity
• Cycle frequency α = Δf: Pilot subcarrier spacing
• Frequency f = nΔf: Individual subcarrier locations These peaks form a characteristic diamond-shaped pattern in the bifrequency plane, enabling blind parameter estimation without prior demodulation.

Stationary Gaussian noise exhibits zero cyclostationarity—its spectral correlation function is identically zero for all non-zero cycle frequencies. This fundamental property allows cyclostationary-based detectors to operate at signal-to-noise ratios (SNR) as low as -20 dB. The detection statistic accumulates correlation over multiple symbols, providing a processing gain proportional to the observation time. This makes cyclostationary signatures the gold standard for spectrum sensing in cognitive radio applications.

Different OFDM-based standards exhibit unique cyclostationary fingerprints:

• LTE downlink: CP length variations (normal vs. extended) shift correlation lag
• 5G NR: Scalable numerology changes cycle frequencies
• WiFi (802.11): Short/long preamble training fields create distinct patterns
• DVB-T: Scattered pilot grid produces standard-specific SCD peaks These signatures enable automatic modulation classification (AMC) systems to identify the specific protocol without decoding the signal.

Full SCD computation via the FAM (FFT Accumulation Method) or SSCA (Strip Spectral Correlation Analyzer) requires O(N²) operations. Practical implementations use:

• Focused cycle frequency detection: Compute SCD only at known α values
• Compressive sensing: Exploit sparsity in the cyclic domain
• Deep learning surrogates: CNNs trained to recognize SCD patterns from raw IQ samples These approaches reduce complexity while preserving detection sensitivity for real-time spectrum monitoring applications.