SSCA Algorithm

The SSCA computes the spectral correlation function by processing the signal in overlapping spectral strips. Unlike the FAM algorithm which uses a channelizer, the SSCA directly computes the complex demodulate of the signal for each strip. This approach allows for a variable resolution product, where the spectral frequency resolution and cycle frequency resolution are inherently linked and can be traded off against each other by adjusting the strip width and the number of data points processed.

The SSCA's primary advantage lies in its computational structure. For a given number of total data samples N, the SSCA requires N complex demodulates, each of which is processed by an N-point FFT. The total complexity is on the order of O(N² log N). This makes it particularly efficient when a high cycle frequency resolution is required over a narrow spectral frequency range, or vice-versa, compared to the FAM algorithm which has a fixed resolution product.

A defining characteristic of the SSCA is its adjustable time-frequency resolution product. The algorithm's parameters—specifically the length of the data segment N and the number of strips N'—directly control the trade-off between spectral frequency resolution (Δf) and cycle frequency resolution (Δα). This allows the analyst to tailor the output to the signal of interest, for example, using fine spectral resolution to separate closely spaced carriers while maintaining sufficient cycle resolution to identify symbol rates.

The core of the SSCA involves computing the complex demodulate of the input signal. For each spectral strip centered at frequency f_k, the signal is frequency-shifted by -f_k and then passed through a low-pass filter. The output of this filter is the complex envelope of the signal within that strip. The spectral correlation is then estimated by correlating the FFTs of these complex demodulates. This direct method avoids the channelization step used in the FAM algorithm.

The SSCA is a consistent estimator of the spectral correlation function, meaning its variance decreases as the number of processed samples increases. However, its reliability is directly tied to the cycle leakage inherent in its strip-based processing. The finite length of the low-pass filter used in the complex demodulate step causes energy at one cycle frequency to leak into adjacent cycle frequency bins, a phenomenon that must be carefully managed through windowing and parameter selection to avoid false peaks in the alpha profile.

While both the FAM and SSCA algorithms estimate the SCF, they differ fundamentally in their approach:

• FAM: Uses a channelizer (a bank of bandpass filters) followed by decimation and FFTs. It has a fixed resolution product.
• SSCA: Uses complex demodulates directly. It offers a flexible resolution product.
• Performance: The FAM is generally more efficient for computing the entire SCF over a wide spectral range with moderate cycle resolution. The SSCA excels when high resolution is needed in one dimension at the expense of the other.

The FFT Accumulation Method is the primary alternative to the SSCA for estimating the Spectral Correlation Function (SCF). While the FAM uses a channelizer and short-time FFTs to compute the cyclic spectrum, the SSCA trades a slightly coarser spectral resolution for significantly lower computational complexity in many scenarios. The choice between FAM and SSCA often depends on whether fine cyclic frequency resolution or faster processing is the priority.

The cyclic periodogram is the basic, inconsistent building block from which both the SSCA and FAM are derived. It is computed from the product of two frequency-shifted, finite-time Fourier transforms. The SSCA improves upon the raw cyclic periodogram by introducing time-averaging across multiple strips, transforming an inconsistent estimator into a reliable, consistent estimate of the true spectral correlation function.

An alpha profile is a one-dimensional slice of the spectral correlation function at a fixed spectral frequency, showing the magnitude of correlation across all cyclic frequencies (alpha). The SSCA's output grid naturally lends itself to extracting alpha profiles, which are critical for blind parameter extraction tasks like symbol rate estimation. Peaks in the alpha profile correspond directly to the signal's cyclic features.

A cyclic feature vector is a compact set of discriminative features derived from the cyclic spectrum at specific cycle frequencies. After the SSCA estimates the spectral correlation function, the peak magnitudes at known cyclic frequencies—such as the symbol rate and carrier frequency offset—are concatenated into a vector. This vector serves as the input to a downstream modulation classifier, enabling robust identification even at low signal-to-noise ratios.

The SSCA is fundamentally an estimator of second-order cyclostationarity, the property of a signal whose autocorrelation function is periodic in time. Most digitally modulated signals—including BPSK, QPSK, and QAM—exhibit strong second-order cyclostationary features. The SSCA exploits this by analyzing the correlation between frequency-shifted signal components, making it a powerful tool for blind modulation recognition without requiring higher-order statistics.