Multi-Cycle Detector

Combines cyclostationary features from multiple cyclic frequencies (α) into a single detection statistic. Unlike a single-cycle detector that examines one periodicity, this approach sums or jointly processes the Spectral Correlation Function (SCF) magnitudes at several α values. This fusion dramatically improves detection sensitivity in low Signal-to-Noise Ratio (SNR) environments where a single cyclic feature may be too weak to detect reliably.

The Generalized Likelihood Ratio Test (GLRT) framework provides the theoretical foundation for optimal multi-cycle detection. It formulates detection as a hypothesis test where the test statistic is a weighted sum of cyclic periodogram magnitudes:

• Weights are derived from the asymptotic covariance of cyclic statistics
• Exploits the fact that cyclic features at different α are asymptotically uncorrelated for distinct cycle frequencies
• Achieves the Cramér-Rao lower bound for detection performance under Gaussian noise assumptions

Processing multiple cycle frequencies increases computational load linearly with the number of α values examined. Efficient implementations use:

• FFT Accumulation Method (FAM) for simultaneous estimation across multiple α slices
• Strip Spectral Correlation Analyzer (SSCA) for real-time multi-cycle processing
• Sub-optimal combining rules (equal-gain, selection combining) to reduce complexity
• Adaptive α selection that dynamically chooses the most reliable cyclic frequencies based on estimated SNR

Multi-cycle detectors inherently reject stationary noise and interference because noise lacks cyclostationarity. By correlating frequency-shifted signal components at multiple α values, the detector:

• Suppresses Gaussian noise which has zero cyclic correlation at α ≠ 0
• Discriminates against co-channel interferers with different cyclic signatures
• Maintains performance under fading channels by exploiting diversity across cycle frequencies
• Enables signal-selective detection when the target signal's cyclic frequencies are known a priori

The alpha profile—a one-dimensional slice of the SCF at a fixed spectral frequency f—serves as the primary input for multi-cycle detection. Integration strategies include:

• Sum-of-magnitudes: Simple non-coherent combining of |S_x^α(f)| across α
• Weighted combination: Using estimated coherence magnitudes as reliability weights
• Joint detection: Processing the full alpha profile vector through a matched subspace detector
• Machine learning fusion: Training a classifier on alpha profile features for adaptive detection thresholds

Multi-cycle detectors operate in two distinct modes:

• Blind detection: Scans all possible α values to detect any cyclostationary signal without prior knowledge. Uses the Dandawate-Giannakis test extended to multiple frequencies
• Signal-specific detection: Targets known cyclic signatures (e.g., symbol rate α = 1/T, carrier frequency offsets) for a specific modulation type
• Hybrid approach: Blindly detects presence, then classifies modulation by matching the detected alpha profile against a cyclic signature database

Multi-cycle detectors enable secondary users to reliably detect primary user signals at very low signal-to-noise ratios (SNRs). By jointly processing features at the symbol rate, carrier frequency offset, and chip rate cyclic frequencies, the detector achieves a higher probability of detection for a given false alarm rate than energy detection or single-cycle methods.

• Detects signals below the noise floor where energy detectors fail
• Distinguishes between multiple primary user waveforms in the same band
• Enables dynamic spectrum access in IEEE 802.22 WRAN and TV white space applications

When classifying unknown signals, a multi-cycle detector extracts a cyclic feature vector by sampling the spectral correlation function at multiple cycle frequencies. This vector forms a unique cyclic signature for each modulation type, enabling robust classification even under severe multipath fading and co-channel interference.

• Combines features from second-order and higher-order cyclostationarity
• Resolves ambiguities where two modulations share a single cyclic feature
• Feeds directly into support vector machines or deep neural network classifiers

A multi-cycle FRESH (FREquency-SHift) filter exploits cyclostationarity at multiple cycle frequencies to separate spectrally overlapping signals. Unlike stationary filters, a FRESH filter processes multiple frequency-shifted versions of the input, each weighted by a linear periodically time-varying impulse response.

• Separates co-channel interference when signals have distinct cyclic frequencies
• Recovers weak signals buried under stronger interferers
• Applied in satellite communications and tactical SIGINT for signal isolation

Cyclic MUSIC extends the standard MUSIC algorithm by exploiting signal cyclostationarity at multiple cycle frequencies. This allows the array to resolve more source signals than the number of physical antenna elements by separating signals in the cyclic frequency domain.

• Resolves co-channel signals arriving from different directions
• Selectively estimates the direction of arrival for signals with a specific cyclic signature
• Critical for electronic warfare support and interference mapping in dense emitter environments

Multi-cycle detectors are applied in radio astronomy to detect man-made or natural signals with periodic statistical structure buried deep in cosmic noise. By integrating correlation across multiple known cyclic frequencies, the detector achieves integration gains beyond what is possible with standard radiometry.

• Detects pulsar signals with known rotational periodicities
• Identifies anthropogenic radio frequency interference (RFI) in observatory data
• Exploits induced cyclostationarity from known transmitter pulse patterns

Multi-cycle detection identifies OFDM signals and estimates their parameters by jointly exploiting cyclostationarity induced by the cyclic prefix and pilot subcarrier patterns. The detector correlates features at cycle frequencies corresponding to the symbol duration, guard interval, and pilot spacing.

• Blindly estimates FFT size, cyclic prefix length, and symbol timing
• Distinguishes between LTE, 5G NR, and WiFi waveforms in shared spectrum
• Enables protocol-aware spectrum monitoring without demodulation