Higher-Order Spectral Analysis (HOSA)

The defining advantage of HOSA is its theoretical insensitivity to Gaussian processes. All higher-order cumulants (order > 2) of a Gaussian distribution are identically zero. This means that when computing the bispectrum or trispectrum of a received signal, additive white Gaussian noise (AWGN) is mathematically eliminated, leaving only the non-Gaussian signal of interest. This property allows for feature extraction at signal-to-noise ratios (SNRs) well below the noise floor where traditional power spectrum analysis fails.

Standard second-order statistics, such as the power spectrum or autocorrelation, are phase-blind. They discard all phase relationships between frequency components. HOSA, specifically the bispectrum, preserves Fourier phase information, making it uniquely capable of detecting Quadratic Phase Coupling (QPC). This occurs when two frequencies interact non-linearly to generate a third, a hallmark of amplifier distortion and a critical hardware fingerprint invisible to conventional spectral analysis.

HOSA provides the mathematical tools to measure deviations from Gaussianity, which are the source of device-specific fingerprints. Key metrics include:

• Skewness (3rd-order): Measures amplitude distribution asymmetry, revealing directional biases in amplifier non-linearity.
• Kurtosis (4th-order): Measures the 'tailedness' of the distribution; excess kurtosis indicates impulsive noise or specific hardware distortion profiles.
• Bicoherence: A normalized bispectrum value (0 to 1) that quantifies the proportion of energy at a bifrequency pair that is quadratically phase-coupled.

Linear systems produce Gaussian outputs from Gaussian inputs. The presence of non-zero higher-order spectra is a definitive indicator of non-linear system behavior. In RF transmitters, non-linearities arise from power amplifiers operating near saturation, mixer intermodulation, and DAC quantization errors. HOSA characterizes these non-linearities through the system transfer function's Volterra series, linking specific bispectral patterns to physical hardware impairments.

The full bispectrum is a two-dimensional function, and the trispectrum is three-dimensional, creating high-dimensional feature spaces. HOSA employs several techniques to compress this information into manageable feature vectors:

• Diagonal Slice Spectrum: A 1-D projection of the bispectrum along its diagonal axis.
• Integrated Polyspectrum: Radial or axial integration of polyspectral values.
• Higher-Order SVD (HOSVD): Multi-linear tensor decomposition that extracts the most discriminative non-Gaussian subspaces for classification.

HOSA is the mathematical backbone of Independent Component Analysis (ICA) for separating co-channel emitters. The central limit theorem dictates that mixtures of independent signals become more Gaussian than their sources. ICA algorithms, such as Joint Approximate Diagonalization of Eigenmatrices (JADE), iteratively maximize non-Gaussianity (measured by kurtosis or negentropy) to recover the original source signals without prior knowledge of the mixing channel.

The third-order frequency-domain representation that detects quadratic phase coupling between signal components. Unlike the power spectrum, the bispectrum preserves phase information, revealing non-linear interactions generated by transmitter hardware impairments. It is defined as the 2D Fourier transform of the third-order cumulant sequence and is theoretically zero for Gaussian noise, making it a powerful tool for Gaussian noise suppression.

The fourth-order frequency-domain representation capturing cubic phase couplings between frequency triplets. It provides a more complete statistical characterization than the bispectrum by detecting interactions invisible to third-order analysis. The trispectrum is essential for identifying subtle non-linearities in complex modulation schemes and offers redundancy that improves robustness in low signal-to-noise ratio conditions.

Statistical measures beyond second-order variance that quantify deviations from Gaussianity. Key cumulants include:

• Skewness (3rd order): Asymmetry in amplitude distribution
• Kurtosis (4th order): Tailedness of the distribution These form the mathematical foundation for constructing polyspectra and are theoretically insensitive to additive Gaussian noise, enabling extraction of hardware-specific fingerprints buried below the noise floor.

A normalized bispectrum that measures the proportion of signal energy at a bifrequency pair that is quadratically phase-coupled. Bounded between 0 and 1, bicoherence provides a quantitative metric for non-linearity detection independent of signal amplitude. Values near 1 indicate strong phase coupling characteristic of specific transmitter non-linearities, while values near 0 suggest random phase relationships or Gaussian noise.

A non-linear phenomenon where two frequency components (f1, f2) interact to generate a third at their sum or difference frequency (f1±f2) with a consistent phase relationship. This coupling arises from amplifier non-linearities and mixer imperfections in transmitter hardware, serving as a distinctive, unclonable fingerprint. HOSA detects this coupling through bispectral analysis, distinguishing genuine hardware interactions from coincidental spectral peaks.

A multi-dimensional array organizing higher-order cumulants that enables joint blind source separation and feature extraction. Represented as a 3D (third-order) or 4D (fourth-order) data structure, cumulant tensors capture the full statistical dependencies between signal samples. Tensor decomposition techniques like HOSVD factor these arrays into interpretable components, compressing the rich higher-order information into compact feature vectors for emitter classification.