Phase Noise Fingerprint

The most visually apparent manifestation of phase noise is the skirting or widening of the carrier signal's base in the frequency domain. Instead of a perfect impulse, the carrier appears as a central peak with a characteristic roll-off. The exact slope of this roll-off, measured in dBc/Hz at specific offset frequencies (e.g., 1 kHz, 10 kHz, 100 kHz), forms a unique spectral mask for each oscillator.

• Close-in noise (offsets < 1 kHz) is dominated by PLL loop filter design and reference oscillator stability.
• Far-out noise (offsets > 100 kHz) is dominated by the VCO's internal thermal noise floor.
• The transition knee between these regions is a highly discriminative feature.

The Phase-Locked Loop is a control system that locks the oscillator's output to a stable reference. The loop bandwidth and damping factor of this control loop leave a distinct imprint on the phase noise spectrum. An under-damped PLL will exhibit a characteristic peaking in the phase noise profile near the loop bandwidth frequency, while an over-damped loop will show a sharp, early roll-off.

• The exact frequency and amplitude of this PLL peaking are highly specific to the discrete component values (resistors, capacitors) in the loop filter.
• Manufacturing tolerances in these passive components create measurable variance between devices of the same design.

Beyond the continuous random phase noise floor, real oscillators generate discrete, deterministic spurs. These appear as narrow, distinct peaks in the spectrum, often caused by power supply ripple coupling, digital clock leakage, or PLL reference feedthrough.

• Power supply spurs: Located at the offset frequency of the switching regulator (e.g., 100 kHz) and its harmonics.
• Reference spurs: Located at the offset equal to the PLL's comparison frequency.
• The exact amplitude and frequency constellation of these spurs acts as a unique comb-like identifier for a specific circuit board layout and power distribution network.

While the spectral mask is a frequency-domain view, the Integrated Phase Jitter is a time-domain metric that captures the total phase error energy over a specific bandwidth. It is calculated by integrating the phase noise power over a defined frequency range (e.g., 1 kHz to 10 MHz).

• This single value, expressed in radians or degrees, represents the total random deviation of the signal's zero-crossings from an ideal clock.
• For fingerprinting, the IPJ value over multiple distinct integration bands provides a compact, highly discriminative feature vector that correlates directly with the physical oscillator's quality factor (Q) and noise sources.

Phase noise does not exist in isolation. The transmitter's Power Amplifier (PA) introduces AM-AM and AM-PM distortion. AM-PM conversion is a critical mechanism where amplitude noise on the signal is converted into additional phase noise by the PA's non-linear phase response.

• This creates a composite fingerprint where the baseband oscillator phase noise is uniquely colored by the PA's specific AM-PM transfer function.
• The statistical correlation between the instantaneous signal envelope and the resulting phase error is a powerful, device-specific feature that cannot be easily cloned, as it depends on the physical interaction of two separate analog components.

The phase noise profile is not static; it exhibits a characteristic drift trajectory as the device warms up from a cold start or reacts to environmental changes. The rate of change of the carrier frequency offset and the reshaping of the phase noise skirt as the crystal oscillator and VCO reach thermal equilibrium form a dynamic signature.

• The thermal time constant of the crystal oven (if present) or the bare crystal creates a predictable warm-up curve.
• A classifier can use this transient drift pattern, observed over the first few seconds of transmission, as an additional layer of identification, distinguishing a device that is merely cold from a different device altogether.

The process of uniquely identifying a specific physical radio transmitter by analyzing the distinct, unintentional features embedded in its emitted waveform. SEI is independent of the encoded data or modulation scheme, relying instead on hardware-intrinsic impairments like phase noise, I/Q imbalance, and power amplifier non-linearity to create a unique fingerprint.

A biometric-like profile of a wireless device constructed from the aggregate of its unique, hardware-intrinsic signal imperfections. RF-DNA combines multiple features—including oscillator phase noise, transient turn-on signatures, and spectral regrowth patterns—into a composite identity that is extremely difficult to clone or spoof, providing a robust foundation for physical-layer authentication.

A higher-order statistical signal processing technique that transforms a signal into the frequency domain to extract features invariant to Gaussian noise. Bispectrum analysis captures the non-linear phase couplings characteristic of specific hardware impairments, making it particularly effective for extracting stable phase noise fingerprints from transmitters operating in low signal-to-noise ratio environments.

An adaptive machine learning mechanism that updates a device's stored fingerprint model over time to account for gradual, environmentally-induced changes. Key factors include:

• Temperature variation affecting oscillator stability
• Component aging shifting phase noise profiles
• Voltage fluctuations altering amplifier characteristics

This ensures long-term authentication reliability without requiring frequent re-enrollment.

A zero-trust security model where a device's physical-layer fingerprint is verified persistently throughout a communication session, rather than only at the initial login. By continuously monitoring the phase noise fingerprint and other hardware signatures, the system can detect session hijacking or device substitution in real-time, triggering immediate revocation of access.

A deep learning training methodology that learns a discriminative embedding space for RF fingerprints. The model is trained to:

• Pull together representations of signals from the same device
• Push apart representations from different devices

This approach is highly effective for phase noise fingerprinting, as it learns to amplify subtle, device-specific phase noise patterns while suppressing channel and noise variations.