Phase Noise

The shape and slope of the phase noise sidebands form a unique spectral signature that varies between oscillators due to manufacturing imperfections. Key aspects include:

• Close-in phase noise (1/f³ and 1/f² regions) dominated by flicker noise in the active devices and resonator Q-factor
• Noise floor (white phase noise region) set by thermal noise and amplifier gain
• The transition frequencies between these regions create a device-specific knee point that is highly repeatable and difficult to clone

Phase noise originates from multiple physical mechanisms within the oscillator circuit:

• Thermal noise in the resonator's motional resistance and sustaining amplifier
• Flicker noise (1/f) upconverted around the carrier by non-linear active devices
• Power supply ripple coupling through finite PSRR, modulating the oscillation frequency
• Vibration-induced noise from microphonic resonators Each mechanism contributes a distinct spectral slope, creating a multi-region fingerprint that reveals the oscillator's architecture and manufacturing quality.

Leeson's heuristic model predicts phase noise behavior based on oscillator parameters:

• L(fm) ∝ (FkT/Ps) × [1 + (f0/2QLfm)²] × (1 + fc/fm)
• Where QL is the loaded Q of the resonator, Ps is the signal power, and F is the amplifier noise figure
• Device-to-device variations in these parameters—especially QL due to manufacturing tolerances—create statistically unique phase noise profiles
• Modern CAD tools use harmonic balance simulations to predict these signatures for fingerprinting model training

Phase noise is particularly valuable for physical layer authentication because:

• It is unclonable—determined by physical manufacturing variances, not programmable parameters
• It persists regardless of modulation format or data content
• It can be extracted using blind estimation techniques without demodulation
• The spectral skirt is robust against multipath fading since it is a relative power measurement near the carrier
• Ensemble methods combining phase noise with I/Q imbalance and PA non-linearity achieve >99% identification accuracy

Extracting phase noise for fingerprinting requires specialized signal processing:

• Direct spectrum analysis using high-dynamic-range instruments with cross-correlation to cancel internal noise
• Delay-line discriminators that convert frequency fluctuations to voltage fluctuations for real-time extraction
• Digital phase-locked loop (DPLL) architectures in SDRs that track and output the instantaneous phase error
• Cyclostationary processing that exploits the periodicity of modulated signals to separate phase noise from modulation
• Residual phase noise after ideal demodulation reveals the transmitter's intrinsic instability

Phase noise signatures drift with environmental conditions, requiring compensation for reliable long-term fingerprinting:

• Temperature shifts alter resonator Q and amplifier bias points, shifting the noise knee
• Aging of crystal resonators causes slow frequency drift and phase noise degradation
• Supply voltage variations modulate oscillator amplitude and phase
• Compensation strategies include polynomial drift models calibrated during enrollment and adaptive tracking filters that update the reference signature
• Multi-sensor fusion with on-board temperature and voltage monitors improves stability

The time-domain counterpart of phase noise, representing the cycle-to-cycle timing uncertainty of a clock edge. In RF fingerprinting, clock jitter directly modulates the sampling instant of ADCs and DACs, imprinting a unique, hardware-specific timing signature onto the digitized waveform.

• Random jitter follows a Gaussian distribution and originates from thermal noise
• Deterministic jitter includes periodic components from power supply coupling
• RMS jitter integrates phase noise over a specific bandwidth

The conversion between phase noise and jitter is mathematically defined, making them two views of the same physical phenomenon.

The root cause of phase noise, stemming from intrinsic device noise in the oscillator's active components and resonator. Different oscillator architectures—such as LC-tank, ring, and crystal—exhibit distinct phase noise profiles that serve as manufacturer-level identifiers.

• Leeson's equation models the phase noise spectrum of feedback oscillators
• Flicker noise (1/f) dominates close-in phase noise below 10 kHz offset
• Thermal noise sets the far-out noise floor
• Quality factor (Q) of the resonator inversely scales phase noise

Temperature-compensated (TCXO) and oven-controlled (OCXO) oscillators each leave unique thermal signatures in their phase noise skirts.

A statistical metric for quantifying frequency stability in oscillators over different observation intervals. Unlike standard deviation, Allan variance converges for common noise types and can discriminate between white frequency noise, flicker frequency noise, and random walk frequency modulation.

• Overlapping Allan deviation provides better confidence for limited datasets
• Modified Allan variance distinguishes white and flicker phase noise
• Hadamard variance handles linear frequency drift

In fingerprinting, the Allan variance curve—plotting stability versus averaging time—creates a multi-scale signature that captures both short-term phase noise and long-term drift characteristics unique to each oscillator.

A feedback control system that synchronizes an oscillator's phase to a reference signal, widely used for frequency synthesis in transmitters. The PLL's loop filter, charge pump, and voltage-controlled oscillator (VCO) each contribute distinct phase noise signatures that propagate to the transmitted signal.

• In-band phase noise is dominated by reference and charge pump noise
• Out-of-band phase noise follows the free-running VCO profile
• Loop bandwidth marks the transition point between these regions
• Fractional-N spurs create deterministic, architecture-specific artifacts

The PLL's locking behavior and transient response to channel switching provide additional fingerprinting features beyond steady-state phase noise.

A phenomenon where phase noise on the local oscillator downconverts strong out-of-band interferers into the desired signal band, degrading the receiver's effective selectivity. In fingerprinting, reciprocal mixing products reveal the oscillator's phase noise profile convolved with the spectral environment.

• Strong blockers near the desired channel are most problematic
• The reciprocal mixing dynamic range quantifies this impairment
• Phase noise at the blocker's offset frequency directly sets the noise floor elevation

This effect can be exploited passively: an observer can characterize a transmitter's oscillator phase noise by analyzing how its signal interacts with known interferers in a dense spectral environment.

Techniques for quantifying single-sideband phase noise include direct spectrum analysis, phase detector methods, and cross-correlation. Each method has trade-offs in sensitivity, offset frequency range, and immunity to amplitude noise that affect fingerprint extraction quality.

• Direct spectrum method uses a spectrum analyzer but has limited dynamic range
• Phase detector method uses a double-balanced mixer to cancel the carrier
• Cross-correlation uses two independent measurement channels to push the noise floor below thermal limits
• Residual phase noise measurements isolate the DUT from reference oscillator contributions

Modern signal source analyzers achieve noise floors below -190 dBc/Hz, enabling extraction of even subtle device-specific phase noise signatures.