Quadratic Phase Coupling

QPC arises from second-order non-linearities in transmitter components, such as power amplifiers operating near saturation. When a signal containing frequencies f1 and f2 passes through a non-linear device, the output includes not only the original frequencies but also intermodulation products.

• Sum frequency generation: f3 = f1 + f2 with phase φ3 = φ1 + φ2
• Difference frequency generation: f3 = f1 - f2 with phase φ3 = φ1 - φ2
• This phase relationship is deterministic and directly tied to the physical hardware path, unlike random noise.
• The coupling strength quantifies the degree of non-linearity, which varies uniquely between devices due to manufacturing variances in analog components.

The bispectrum is the primary mathematical tool for detecting and quantifying QPC. It is the Fourier transform of the third-order cumulant and measures the statistical dependence between three frequency components.

• A non-zero bispectrum value at bifrequency (f1, f2) indicates that the energy at f3 = f1 + f2 is phase-coupled to f1 and f2.
• Gaussian noise suppression: The bispectrum of a Gaussian process is theoretically zero, making QPC detection immune to Gaussian background noise.
• The bicoherence, a normalized bispectrum, provides a bounded measure (0 to 1) of the fraction of power at f3 that is quadratically coupled.
• This allows engineers to distinguish between spontaneously generated harmonics and true non-linear coupling.

QPC patterns are highly device-specific because they originate from the unique physical layout and semiconductor properties of each transmitter's analog chain.

• Power amplifier non-linearity: The AM-AM and AM-PM distortion curves of an amplifier directly shape the QPC signature.
• Mixer imperfections: Local oscillator leakage and port-to-port isolation in mixers create distinct coupling patterns.
• PCB trace interactions: Parasitic capacitances and inductances on the circuit board introduce subtle, repeatable non-linearities.
• These features are unclonable—even with an identical make and model, an adversary cannot replicate the exact QPC signature because it depends on atomic-level manufacturing variances.

A critical advantage of QPC analysis for RF fingerprinting is its theoretical immunity to additive Gaussian noise. Standard power spectrum analysis can be easily buried by noise, but higher-order statistics exploit a fundamental property.

• The third-order cumulant (and thus the bispectrum) of any Gaussian process is identically zero.
• This means that even when a signal is deeply buried below the noise floor, its QPC signature remains detectable.
• Practical benefit: Emitter identification can be performed at significantly lower signal-to-noise ratios (SNR) than with second-order methods.
• This is particularly valuable in electronic warfare and spectrum surveillance scenarios where signals are intentionally weak or distant.

The full bispectrum is a two-dimensional function, which can be computationally expensive and data-intensive. The diagonal slice is a practical simplification that captures essential QPC information along a single axis.

• Defined as B(f, f), the diagonal slice measures coupling where f1 = f2, generating the second harmonic 2f.
• It reduces the 2D bispectrum to a 1D feature vector, dramatically lowering computational complexity.
• Integrated bispectrum extends this concept by radially integrating the bispectrum along paths, condensing information while preserving discriminative power.
• These techniques make QPC-based fingerprinting feasible for real-time, embedded edge AI systems with limited memory and processing.

Not every peak at a sum frequency indicates QPC. It is essential to distinguish true phase-coupled components from spontaneously generated harmonics that lack a consistent phase relationship.

• A harmonic generated by a simple non-linearity will have a random or independent phase relative to the fundamentals.
• Bicoherence analysis makes this distinction explicit: a bicoherence value near 1 indicates strong QPC, while a value near 0 indicates an independent, uncoupled component.
• This prevents false positives in fingerprinting where a generic distortion product might be mistaken for a unique device signature.
• Only the phase-coupled energy contributes to a stable, repeatable fingerprint suitable for machine learning classification.