Phase-Noise Power Spectral Density (PN-PSD)

• PN-PSD is a statistical measure that characterizes phase fluctuations using Fourier frequency analysis, clearly distinguishing between white and colored noise.
• It enables accurate modeling and prediction of system performance in domains like communications, metrology, and astronomical imaging by linking frequency and time-domain metrics.
• Advanced extraction methods, including periodogram techniques and Bayesian regression, facilitate reliable estimation and compensation of composite noise components.

Phase-Noise Power Spectral Density (PN-PSD) quantifies the distribution of phase fluctuations in oscillators, lasers, and optical wavefronts as a function of Fourier frequency. PN-PSD provides a frequency-resolved statistical characterization of phase noise, enabling direct links to time-domain stability metrics (e.g., Allan variance), quantitative models of practical oscillator impairments, and predictive analysis of system performance in metrology, communications, and astronomical imaging. Unlike the traditional “linewidth” metric, PN-PSD distinguishes between white and colored noise contributions, facilitates modeling of composite and power-law processes, and underpins both the formulation of accurate physical models and the design of optimal estimation and compensation algorithms in diverse domains.

1. Formal Definition and Physical Interpretation

The phase-noise PSD $S_{\phi}(f)$ is formally defined as the one-sided Fourier transform of the autocorrelation of the phase-fluctuation process $\phi(t)$ (assumed zero-mean and stationary):

$$S_{\phi}(f) = 2 \lim_{T \to \infty} \frac{\mathbb{E}[|\tilde{\phi}_T(f)|^2]}{T}$$

where $\tilde{\phi}_T(f)$ is the finite-time Fourier transform: $$\tilde{\phi}_T(f) = \int_{-T/2}^{T/2} \phi(t) e^{-i 2\pi f t} dt$$ This quantifies the phase-noise power at each offset frequency $f$ from the carrier, with units of $\mathrm{rad^2/Hz}$. In spatially extended systems, such as turbulent optical wavefronts, the phase-variance contribution from spatial frequencies around $\mathbf{f}$ is given by the spatial phase-PSD $W_\phi(\mathbf{f})$, the Fourier transform of the two-point spatial autocorrelation function $B_\phi(\rho)$ (Fétick et al., 2018).

The PN-PSD encapsulates all information required to compute the total phase variance:

$$\mathbb{E}[\phi^2(t)] = \int_{0}^{\infty} S_{\phi}(f) df$$

allowing direct linkage to integrated phase jitter, linewidth, and system performance bounds.

2. Power-Law PN-PSD Models and Physical Regimes

Phase noise in practical systems is rarely spectrally white; instead, $S_{\phi}(f)$ is commonly represented as a sum of power laws:

$S_{\phi}(f) = \sum_{\alpha} h_\alpha f^{\alpha}$

with typical exponents $\alpha \in \{0, -1, -2, -3, -4\}$, corresponding respectively to white phase modulation (WPM), flicker phase modulation (FPM), random-walk phase modulation (RWPM), flicker frequency modulation (FFM), and random-walk FM (Georgakaki et al., 2012). Each noise process dominates in specific frequency regions, producing characteristic slopes on log–log $S_{\phi}(f)$ plots.

For example, in communication oscillator models, the single-sideband PN-PSD often displays a $1/f^2$ (Wiener) region at intermediate offsets, flat (white) regions at both low (PLL-locked) and high (thermal or device noise) offset frequencies, and may be parametrized:

$$S_\theta(f) = \frac{10^{10}L_{100}}{\Delta\omega + f^2} + L_\infty$$

with $\Delta\omega$ the 3dB linewidth, $L_{100}$ the SSB phase noise at 100 kHz offset, and $L_\infty$ the high-frequency white phase noise floor (Piemontese et al., 2021).

In atmospheric wavefronts, the PSD may follow the Kolmogorov spectrum:

$W_\phi(f) = 0.023 r_0^{-5/3} f^{-11/3}$

or the von Kármán spectrum (with outer scale $L_0$)

$$W_\phi(f) = 0.023 r_0^{-5/3}(f^2 + L_0^{-2})^{-11/6}$$

Suitable for modeling turbulent aberrations in large apertures (Fétick et al., 2018).

3. Extraction and Estimation of PN-PSD

PN-PSD estimation from time- or space-domain data employs periodogram methods, windowing, and advanced Bayesian techniques. The classic one-sided periodogram estimator for $N$ samples $\phi_n$ at sample rate $f_s$ is

$$\hat S_{\phi}(f_k) = \frac{2}{N f_s} \left| \sum_{n=0}^{N-1} \phi_n e^{-i2\pi f_k n \tau_0} \right|^2$$

where $f_k = k / (N\tau_0), \; k=0,\ldots,N/2$ (Georgakaki et al., 2012). Window-averaging and log–log linear regression extract power-law slopes and intercepts.

Bayesian estimators, such as the Bretthorst periodogram, are used when model assumptions (e.g., stationarity, whiteness) break down or to resolve oscillatory components.

Where only time-domain stability metrics such as Allan variance $\sigma_y^2(\tau)$ are available, transformation algorithms can generate an approximate multi-segment power-law PN-PSD (Marchi et al., 2023). The forward mapping from PSD to Allan variance is unique; the inverse, for composite noise types, is only approximate, requiring careful piecewise fitting.

4. PN-PSD in Physical and Engineering Contexts

a. Optical and Photonic Systems

The PN-PSD directly determines the statistical behavior of phase-jitter in mode-locked lasers, comb generation, four-wave mixing, and interferometric setups. It governs the degree of coherence, linewidth broadening, and ultimate phase-stability of frequency combs and soliton molecules (Tian et al., 2020). In nonlinear optics, the FM-noise PSD $S_F(f)$, related to $S_{\varphi}(f)$ by $S_F(f) = (2\pi f)^2 S_{\varphi}(f)$, governs how phase noise propagates through nonlinear processes, as in four-wave mixing, where output noise PSDs are affine combinations (e.g., $$S_{F,\mathrm{Stokes}} = 4 S_{F,\mathrm{pump}} + S_{F,\mathrm{signal}}$$) (Anthur et al., 2013).

b. Frequency Metrology and Oscillator Characterization

PN-PSD measurement is fundamental for specifying oscillator quality, timekeeping performance, and design of frequency synthesizers. An explicit PN-PSD model allows stochastic simulation and direct translation between frequency- and time-domain stability metrics (Marchi et al., 2023). It is also indispensable for analyzing colored and multi-regime noise, quantifying measurement uncertainty in the presence of non-white noise, and designing metrological experiments (Georgakaki et al., 2012).

c. Communication Systems

In communications, PN-PSD impacts carrier synchronization, residual phase error, power loss, and inter-symbol interference (ISI). Closed-form expressions relate measured PN-PSD parameters to discrete-time phase-noise models (e.g., AR(1) processes matched to measured variance), power-loss, and optimal symbol rates for communication under phase noise (Piemontese et al., 2021). In MIMO systems and SDR hardware, empirical multi-parameter PN-PSD models are constructed by log–log regression to accurately simulate link-layer phase impairments (Collmann et al., 16 Jul 2025).

d. Astronomical Imaging

Wavefront PN-PSD under atmospheric turbulence, and its modification by adaptive optics (AO), fundamentally sets the long-exposure point-spread function (PSF) and imaging halo profiles. The analytical expansion of the PSF as a series in convolutive orders of the phase PSD enables efficient simulation, AO diagnostics, and inversion for atmospheric parameters (Fétick et al., 2018).

5. Experimental and Algorithmic Compensation in PN-PSD Measurement

Experimental PN-PSD characterization faces artifacts induced by finite delay (e.g., fringe notches in short-delay self-heterodyne setups), noise-floor contamination, and gaps ("poles") at frequencies where transfer functions vanish (Riebesehl et al., 21 May 2025). Advanced digital signal processing methods integrate kernel-based regression (e.g., kernel ridge regression using an RBF kernel) with power-spectrum equalization (PSE) frameworks to automatically reconstruct the true PN-PSD, using log-space surrogates trained on high-SNR spectral regions and robust cross-validation for hyperparameter tuning. Such methods enable artifact-free PN-PSD estimates with sub-dB accuracy over the entire frequency band, even for complex or multi-slope lineshapes.

6. Time-Domain Links: Allan/Modified Allan Variance

Direct links exist between PN-PSD and time-domain stability measures:

• The frequency-deviation PSD, $S_y(f) = f^2 S_\phi(f)$, allows computation of Allan variance: $$\sigma_y^2(\tau) = 2 \int_0^{\infty} S_y(f) \frac{ \sin^4(\pi f \tau) }{ (\pi f \tau )^2 } df$$
• Modified Allan variance, involving $\sin^6(\pi f \tau )$ kernel, uniquely differentiates between white and flicker PM noise, resolving ambiguities in oscillator characterization (Georgakaki et al., 2012).

Approximating PSD from AVAR involves:

1. Fitting $\sigma_y^2(\tau)$ as piecewise power laws to identify noise regimes.
2. Mapping exponents between time and frequency domain ($\alpha = -\mu -1$).
3. Using closed-form evaluations and continuity constraints to assemble a composite PSD model (Marchi et al., 2023).

7. PN-PSD in System Modeling and Simulation

Closed-form, multi-parameter models for PN-PSD can be systematically estimated from measured data (e.g., by identifying plateau and slope regions in phase-locked loop (PLL) oscillators). These models encapsulate key underlying physical processes—flicker and white FM, loop transmission, and noise-floor transitions—and support realistic generation of synthetic phase-noise time series for Monte Carlo or system-level simulation in both communications and metrological applications (Collmann et al., 16 Jul 2025, Marchi et al., 2023).

Efficient convolutional-order expansions of PSF in terms of PN-PSD under spatial turbulence enable rapid, physically interpretable simulation tools for telescope/instrument design and AO system diagnostics (Fétick et al., 2018).

---

In summary, the PN-PSD is a central tool in quantitative phase noise analysis across optical/quantum, electronic, and astronomical domains, providing both a universal language for noise description and a rigorous engine for propagation, mitigation, and exploitation of phase fluctuations in advanced measurement and communication systems.