Cyclic Signature

Each modulation scheme generates a distinct pattern of cyclic frequencies (α) and associated spectral correlation magnitudes. This pattern acts as a unique identifier, allowing classifiers to distinguish between signals that appear identical in the power spectrum.

• BPSK exhibits strong cyclostationarity at α = 2fc and α = 2fc ± symbol rate
• QPSK lacks the doubled carrier feature but shows strong correlation at the symbol rate
• 16-QAM displays a more complex profile with features at multiple cyclic frequencies related to its constellation geometry

Cyclic signatures are inherently robust against stationary noise and interference. Because thermal noise is a stationary process, it exhibits no cyclostationarity and contributes zero energy to the cyclic spectrum at non-zero cycle frequencies.

• Stationary Gaussian noise collapses to α = 0 in the cyclic domain
• The signal's cyclic features remain detectable at signal-to-noise ratios far below the noise floor
• This property makes cyclic analysis superior to energy detection for spectrum sensing in contested environments

The cyclic signature directly reveals a signal's physical-layer parameters without prior knowledge or demodulation. Key parameters are extracted by identifying peaks in the alpha profile:

• Symbol rate: Detected as a strong cyclic feature at α = Rs
• Carrier frequency offset: Estimated from the location of doubled-carrier features
• Pulse shape roll-off: The width of cyclic features around the symbol rate reveals the excess bandwidth
• OFDM guard interval: The cyclic prefix induces a distinct feature at α = 1/Ts

Cyclic signatures can be inherent to the modulation process or intentionally induced for identification purposes.

Inherent sources:

• Symbol rate periodicity from pulse shaping
• Carrier frequency doubling in BPSK
• Pilot tones and training sequences

Induced sources:

• Deliberate insertion of periodic preambles
• Transmitter-specific pulse shaping variations
• Intentional spectral correlation for RF watermarking

Induced features enable physical-layer authentication and transmitter identification.

Combining information from multiple cyclic frequencies dramatically improves detection and classification performance. A multi-cycle detector integrates the spectral correlation at several known α values.

• Single-cycle detection may miss signals in deep fading
• Multi-cycle integration provides diversity gain against channel impairments
• The Dandawate-Giannakis test provides a statistically optimal framework for multi-cycle detection
• Performance approaches the theoretical optimum as more cycle frequencies are incorporated

Two primary algorithms compute the cyclic signature from sampled IQ data:

FAM (FFT Accumulation Method):

• Uses a channelizer and short-time FFTs
• Efficient for wideband signals with many cyclic features
• Complexity: O(N² log N)

SSCA (Strip Spectral Correlation Analyzer):

• Processes the signal in spectral strips
• Better resolution at low cycle frequencies
• Different memory-compute trade-offs

Both produce the cyclic domain profile used as input to neural network classifiers.

The foundational two-dimensional transform that defines a cyclic signature. It measures the correlation between frequency-shifted versions of a signal, revealing hidden periodicities in its spectral content. The SCF is a function of both spectral frequency (f) and cyclic frequency (α), and its non-zero magnitudes at specific α values form the unique pattern that characterizes a modulation scheme.

A compact, machine-readable representation of a cyclic signature used as direct input to a modulation classifier. It is constructed by extracting the magnitudes of the SCF or cyclic cumulants at a pre-selected set of discriminative cyclic frequencies (α). This vector reduces the dimensionality of the full cyclic spectrum while preserving the unique identifying features of the modulation scheme.

A one-dimensional slice of the cyclic signature obtained by fixing the spectral frequency (f) and plotting the magnitude of spectral correlation across all cyclic frequencies (α). Peaks in the alpha profile directly correspond to the fundamental periodicity of the signal, such as the symbol rate and carrier frequency offset, making it a critical tool for blind parameter extraction.

A comprehensive visualization of the entire cyclic signature, typically rendered as a 2D plot with cycle frequency (α) on one axis and spectral frequency (f) on the other. The pattern of correlation peaks in this domain acts as a unique visual fingerprint for a modulation type, revealing features like the presence of a cyclic prefix in OFDM or the chip rate in spread-spectrum signals.

A technique for intentionally engineering a cyclic signature at the transmitter to simplify identification. By inserting a known periodic pattern, such as a specific pilot sequence or a unique pulse-shaping filter, a controlled cyclostationary feature is created. This allows a receiver to detect and classify the signal by simply correlating against the known induced cyclic frequency, bypassing complex blind estimation.

A scalar metric that quantifies the strength of a cyclic signature relative to the signal's stationary power. It provides a single numerical value indicating how pronounced the cyclostationary features are. A high degree of cyclostationarity implies a robust signature that is easily detectable in noise, while a low degree suggests a weak signature that may be masked by interference.