Cyclostationary Feature Detection

The Spectral Correlation Function (SCF) is the fundamental mathematical transform that reveals cyclostationarity. It measures the correlation between two frequency-shifted versions of a signal. For a cyclostationary signal, the SCF produces non-zero values at specific cyclic frequencies (α) and spectral frequencies (f). Stationary noise, lacking this periodic structure, exhibits zero correlation for α ≠ 0. This allows the detector to distinguish a modulated signal from background noise even at SNR values below -20 dB.

Each modulation scheme generates a unique cyclostationary fingerprint. Key examples include:

• BPSK: Strong cyclic features at symbol rate and twice the carrier offset
• QPSK/OQPSK: Features at symbol rate; OQPSK lacks the doubled carrier feature
• OFDM: Cyclic prefix induces features at the OFDM symbol rate and its multiples
• GMSK: Exhibits features at the symbol rate and specific multiples tied to the Gaussian filter This enables automatic modulation classification without prior knowledge of the signal.

The core advantage of cyclostationary detection is its inherent immunity to stationary Gaussian noise. Thermal noise is wide-sense stationary and exhibits no spectral correlation for non-zero cyclic frequencies. By computing the SCF at α ≠ 0, the detector mathematically nullifies the noise contribution entirely. This contrasts sharply with energy detection, which suffers from a noise uncertainty floor and fails completely below a known SNR wall. Cyclostationary detectors have no such theoretical floor.

Beyond simple detection, cyclostationary analysis directly estimates key physical-layer parameters:

• Symbol Rate: Identified as the cyclic frequency of the strongest feature
• Carrier Frequency Offset: Derived from the location of conjugate cyclic features
• Guard Interval Length: Extracted from the cyclic autocorrelation of the cyclic prefix in OFDM signals
• Chip Rate: Determined for direct-sequence spread spectrum signals This provides a complete blind signal characterization capability for cognitive radios.

Full SCF computation is computationally intensive. Practical implementations use optimized algorithms:

• FFT Accumulation Method (FAM): A time-smoothing approach that reduces complexity from O(N²) to O(N log N) by using a channelizer and decimation
• Strip Spectral Correlation Analyzer (SSCA): A frequency-smoothing method optimized for real-time, pipelined hardware implementation on FPGAs
• Compressive Cyclostationary Detection: Applies compressed sensing to sample the SCF at sub-Nyquist rates, dramatically reducing ADC and processing requirements for wideband monitoring.

Cyclostationary detectors can simultaneously detect and separate multiple overlapping signals in the same frequency band. Because each signal's cyclic features are a function of its unique symbol rate, carrier offset, and modulation format, the SCF domain naturally separates signals that overlap in both time and frequency. This enables co-channel signal separation and interference identification without requiring spatial filtering or beamforming, a critical capability for spectrum enforcement and electronic warfare applications.

The fundamental mathematical function computed in cyclostationary detection. It measures the correlation between spectral components of a signal separated by a cyclic frequency (α). Unlike the Power Spectral Density (PSD), which treats a signal as stationary, the SCD reveals hidden periodicities by analyzing the signal in the bi-frequency plane. A signal exhibits cyclostationarity if its SCD shows non-zero values for α ≠ 0, allowing it to be distinguished from stationary noise, which has no spectral correlation.

The time-domain equivalent of the Spectral Correlation Density. The CAF computes the correlation of a signal with a frequency-shifted and conjugated version of itself. Key properties exploited for detection:

• Quadratic transformation: Converts a cyclostationary signal into a sine wave at the cyclic frequency, making it detectable with a narrowband filter
• Noise rejection: Stationary white noise has a CAF of zero for non-zero cyclic frequencies
• Modulation fingerprinting: Different modulation schemes (BPSK, QPSK, OFDM) produce distinct CAF patterns

A simplified form of cyclostationary detection specifically for OFDM signals (used in LTE, 5G, Wi-Fi). The cyclic prefix is a copy of the end of each OFDM symbol appended to its beginning. This intentional redundancy creates a strong cyclostationary signature at the symbol rate. Detection is performed by computing the autocorrelation with a lag equal to the useful symbol duration, requiring significantly less computational complexity than full SCD estimation while remaining effective at very low SNRs.

A critical distinction for spectrum sensing design:

• Wide-Sense Stationary (WSS): Mean and autocorrelation are time-invariant. Noise and unstructured interference fall here
• Cyclostationary: Statistical properties vary periodically with time. All modulated signals exhibit this due to carrier frequency, symbol rate, or guard intervals
• Why it matters: Energy detectors cannot distinguish between a modulated signal and high-power noise. Cyclostationary detectors exploit the periodicity unique to man-made signals, enabling detection at SNR levels as low as -20 dB.

The most widely implemented algorithm for efficient SCD estimation. The FAM reduces computational complexity from O(N³) to O(N² log N) by:

• Channelization: Using a sliding FFT to decompose the input into narrowband components
• Decimation: Downsampling each channel to reduce the data rate
• Cross-correlation: Computing the spectral correlation between frequency-shifted channel pairs This makes real-time cyclostationary detection feasible on FPGA and GPU hardware for wideband spectrum monitoring applications.

Beyond simple signal detection, cyclostationary analysis enables blind modulation classification. Each modulation format generates a unique cyclic signature:

• BPSK: Strong feature at α = 2fc (carrier harmonic) and α = symbol rate
• QPSK: Feature at α = symbol rate only; no carrier harmonic
• MSK/GMSK: Features at α = 2fc ± symbol rate/2
• OFDM: Features at α = symbol rate and cyclic prefix rate This allows cognitive radios to identify the transmission type of primary users without demodulation.