Fringe Frequency in Interferometry

• Fringe Frequency is a family resemblance concept with definitions varying by domain, including spatial, temporal, and spectral interpretations across interferometry and holography.
• It encodes critical phase and geometrical information that drives design choices, sensitivity, and phase unwrapping in applications from optical metrology to gravitational-wave analysis.
• Recent advances leverage deep learning and Fourier methods to accurately reconstruct phase and orientation in fringe pattern analysis, enhancing measurement precision across techniques.

Searching arXiv for recent and foundational papers related to “fringe frequency” across interferometry and fringe-pattern analysis. Fringe frequency is not a single invariant quantity across interferometry and fringe-pattern analysis. In optical spectro-interferometry it can denote the spatial frequency sampled by a baseline, $u=B_\perp/\lambda$ (Armstrong et al., 2019). In radio interferometry it is the temporal Fourier coordinate conjugate to local sidereal time (LST) (Garsden et al., 2024). In fringe projection profilometry it is often expressed through fringe pitch or period number, with $\text{frequency}=W/\lambda$ and larger period number corresponding to higher spatial frequency (Kim et al., 2024). In holographic vibration metrology, the practically useful marker can be sideband order $n$, because significant sideband energy is confined mostly to $|n|\lesssim \Phi$ (Joud et al., 2010). In gravitational-wave lensing and BISER spectroscopy, fringe frequency refers instead to oscillation spacing in frequency or photon-energy space (Jung et al., 2018, Ogura et al., 30 Jun 2025). The common structure is an oscillatory observable whose spacing, rate, or Fourier coordinate encodes geometry, motion, bandwidth, or source structure.

1. Domain-specific meanings and canonical forms

Across the literature, the phrase is used in several technically distinct senses. In some cases it denotes a directly sampled Fourier coordinate; in others it denotes a carrier, a sideband index, or a spectral oscillation spacing. The most useful way to organize the concept is by measurement domain rather than by a single universal definition (Armstrong et al., 2019, Garsden et al., 2024, Kim et al., 2024, Joud et al., 2010, Jung et al., 2018, Ogura et al., 30 Jun 2025).

Domain	Operative meaning	Representative relation
Optical spectro-interferometry	Spatial frequency sampled by a baseline	$u = B_\perp/\lambda$
Radio interferometry	Fringe rate, Fourier pair of LST	$$f \approx \frac{\nu}{c}(\boldsymbol{b}(t)\times \omega_\oplus)\cdot \boldsymbol{n}$$
Fringe projection profilometry	Spatial frequency via pitch or period number	$\text{frequency}=\frac{W}{\lambda}$
Holographic vibration metrology	Sideband order used as local amplitude marker	$I_n(\Phi)=|\mathcal{E}J_n(\Phi)|^2$
Cosmic-string GW lensing	Oscillation spacing in detector frequency	$f_{\rm width}$
BISER spectroscopy	Spectral fringe spacing in $\omega$ or $\text{frequency}=W/\lambda$0	$\text{frequency}=W/\lambda$1

This multiplicity is not terminological looseness so much as a reflection of different measurement operators. A baseline samples a sky Fourier mode; a structured-light projector encodes object shape into spatial carrier phase; a holographic vibration field redistributes power into Bessel sidebands; and a chirped or aperiodic source generates interference fringes in a spectral variable. A plausible implication is that “fringe frequency” is best treated as a family resemblance concept whose precise meaning is fixed by the conjugate domain being analyzed.

2. Holographic and interferometric metrology

In holographic vibration analysis, the classical fringe-frequency question is displaced from spatial fringe counting to sideband localization. For sinusoidal vibration $\text{frequency}=W/\lambda$2, the scattered optical field is phase modulated and expands as

$\text{frequency}=W/\lambda$3

with sideband intensities

$\text{frequency}=W/\lambda$4

Conventional time-averaged holography uses zeros of $\text{frequency}=W/\lambda$5, but the sideband method in "Fringe free holographic measurements of large amplitude vibrations" (Joud et al., 2010) instead examines $\text{frequency}=W/\lambda$6 at fixed pixel and exploits the sharp cutoff near $\text{frequency}=W/\lambda$7. Because significant sideband energy remains mostly confined to $\text{frequency}=W/\lambda$8, the sideband order itself becomes a local amplitude marker, eliminating fringe counting and reference-point comparison. The demonstration on a clarinet reed measured phase modulations up to about $\text{frequency}=W/\lambda$9 radians and displacements of about $n$0, including cases where spatial fringes were unresolved (Joud et al., 2010).

A different use appears in deep frequency modulation interferometry. "Compact multi-fringe interferometry with sub-picometer precision" (Isleif et al., 2019) explicitly notes that it does not define a separate fringe-crossing frequency. Instead, a sinusoidal laser frequency modulation at $n$1 produces harmonics at $n$2, and phase is extracted from a Bessel-function decomposition of about the first ten harmonics. The modulation index is

$n$3

and multi-fringe capability means continuous phase tracking across many $n$4 cycles rather than explicit fringe counting. In this setting, the physically meaningful quantity is the interferometric phase encoded across harmonic amplitudes and phases, not a separately tabulated fringe frequency (Isleif et al., 2019).

These two cases show a recurrent distinction. In holographic vibration metrology, fringe information is profitably reinterpreted as a frequency-sideband support problem. In DFMI, fringe information is encoded in a harmonic comb generated by deep modulation. In both, the central observable is phase, but the frequency-domain representation is the practically useful one.

3. Fringe projection profilometry and spatial carrier design

In fringe projection profilometry, fringe frequency is directly tied to sensitivity and ambiguity. "Enhanced fringe-to-phase framework using deep learning" (Kim et al., 2024) defines

$n$5

with $n$6 the projector width and $n$7 the fringe pitch, so smaller pitch means denser fringes and higher frequency. The same paper uses pitches $n$8, corresponding to frequencies $n$9, and constructs a refined reference phase by combining information from frequencies different from those used as network inputs (Kim et al., 2024). "Self-supervised phase unwrapping in fringe projection profilometry" (Gao et al., 2023) expresses the same trade-off in terms of period number $|n|\lesssim \Phi$0: larger period number means higher spatial fringe frequency and better depth sensitivity, but also more difficult fringe-order recovery. It states that in conventional dual-frequency temporal phase unwrapping the high-frequency pattern is usually limited to about $|n|\lesssim \Phi$1 or lower, whereas its self-supervised method retrieves the absolute fringe order from one $|n|\lesssim \Phi$2-period wrapped phase map and improves average depth RMSE from $|n|\lesssim \Phi$3 to $|n|\lesssim \Phi$4 over $|n|\lesssim \Phi$5 test samples (Gao et al., 2023).

A limiting case is deliberate operation at the Nyquist boundary. "Profilometry with digital fringe-projection at the spatial and temporal Nyquist frequencies" (Padilla et al., 2017) sets

$|n|\lesssim \Phi$6

At spatial Nyquist, the projected sinusoid is sampled with two pixels per period and becomes binary; slight projector defocusing serves as an analog low-pass filter. At temporal Nyquist, two frames differ by a $|n|\lesssim \Phi$7 phase shift, so frame subtraction implements

$|n|\lesssim \Phi$8

which rejects the DC term and all even-order harmonics. After subtraction, a one-dimensional spatial Hilbert filter isolates the analytic component. This is an unusually explicit demonstration that “fringe frequency” can be a design variable pushed to the digital limit rather than merely a nuisance parameter (Padilla et al., 2017).

These formulations all preserve the same principle: increasing spatial carrier frequency increases phase sensitivity, but also strengthens wrapping, harmonic, and sampling constraints. The literature therefore treats frequency choice as a coupled problem involving geometry, unwrapping, demodulation architecture, and runtime frame budget.

4. Fourier limits, deep demodulation, and orientation-based local analysis

The classical single-frame Fourier literature treats fringe frequency as a carrier-separation condition. "Low frequency fringe pattern analysis via a Fourier transform method" (Henning et al., 2018) studies the regime where only a few sinusoidal cycles occur across the record. For

$|n|\lesssim \Phi$9

finite support broadens each sideband through the window transform, causing overlap of positive and negative lobes, DC contamination, and asymmetric loss of desired spectral content. The paper shows that the usual Takeda sideband-separation argument degrades fundamentally when the carrier is low, with the largest phase errors near the record edges. It recommends background subtraction before the Fourier transform and explicit control of the envelope spectrum. Here fringe frequency is the carrier wavenumber $u = B_\perp/\lambda$0, and low carrier means the sideband geometry itself ceases to support accurate analytic-signal recovery (Henning et al., 2018).

A data-driven alternative appears in "Fringe pattern analysis using deep learning" (Feng et al., 2018). That work is experimentally verified on carrier fringe patterns in fringe projection profilometry, with projected spatial frequency $u = B_\perp/\lambda$1, and reports that the learned method remains robust even for a relatively low frequency $u = B_\perp/\lambda$2. By contrast, Fourier transform profilometry suffers from “spectra leakage and overlapping in the frequency domain,” while windowed Fourier profilometry smooths the surface but fails to preserve some details near boundaries and abrupt depth-changing regions (Feng et al., 2018). In this setting, the network does not explicitly estimate fringe frequency; it predicts intermediate quantities $u = B_\perp/\lambda$3, $u = B_\perp/\lambda$4, and $u = B_\perp/\lambda$5, and phase is recovered from $u = B_\perp/\lambda$6 (Feng et al., 2018).

A third viewpoint is directional rather than scalar. "DeepOrientation: convolutional neural network for fringe pattern orientation map estimation" (Cywinska et al., 2023) addresses the local fringe orientation map $u = B_\perp/\lambda$7, not the local frequency magnitude. The paper explicitly lists “local fringe spatial frequency (fringe period) estimation” as a task facilitated by accurate orientation. Orientation is represented via doubled-angle components $u = B_\perp/\lambda$8 and $u = B_\perp/\lambda$9, which avoid modulo-$$f \approx \frac{\nu}{c}(\boldsymbol{b}(t)\times \omega_\oplus)\cdot \boldsymbol{n}$$0 discontinuities and support downstream algorithms such as Hilbert spiral transform demodulation (Cywinska et al., 2023). This suggests a useful decomposition: local fringe analysis may estimate the direction of $$f \approx \frac{\nu}{c}(\boldsymbol{b}(t)\times \omega_\oplus)\cdot \boldsymbol{n}$$1 and the magnitude of $$f \approx \frac{\nu}{c}(\boldsymbol{b}(t)\times \omega_\oplus)\cdot \boldsymbol{n}$$2 separately, with orientation estimation stabilizing the subsequent local frequency estimation problem.

Collectively, these studies divide the subject into two regimes. In one, fringe frequency is an explicit carrier to be separated in Fourier space. In the other, it is partially absorbed into a learned representation that predicts phase-related intermediates or orientation fields without explicit spectral filtering.

5. Astronomical and radio interferometry

In stellar optical interferometry, fringe frequency is often literally the sampled spatial frequency. "Interferometric Fringe Visibility Null as a Function of Spatial Frequency: a Probe of Stellar Atmospheres" (Armstrong et al., 2019) defines

$$f \approx \frac{\nu}{c}(\boldsymbol{b}(t)\times \omega_\oplus)\cdot \boldsymbol{n}$$3

and studies the null $$f \approx \frac{\nu}{c}(\boldsymbol{b}(t)\times \omega_\oplus)\cdot \boldsymbol{n}$$4 where $$f \approx \frac{\nu}{c}(\boldsymbol{b}(t)\times \omega_\oplus)\cdot \boldsymbol{n}$$5. It then parameterizes the null through the equivalent uniform-disk diameter

$$f \approx \frac{\nu}{c}(\boldsymbol{b}(t)\times \omega_\oplus)\cdot \boldsymbol{n}$$6

In the $$f \approx \frac{\nu}{c}(\boldsymbol{b}(t)\times \omega_\oplus)\cdot \boldsymbol{n}$$7 Ophiuchi demonstration, the observed $$f \approx \frac{\nu}{c}(\boldsymbol{b}(t)\times \omega_\oplus)\cdot \boldsymbol{n}$$8 slope required more limb darkening at shorter wavelengths than either of the atmosphere models considered (Armstrong et al., 2019). Here “fringe frequency” is a baseline-sampled Fourier coordinate on the sky, and the visibility null is a direct probe of atmospheric structure.

Finite bandwidth introduces a different fringe-frequency structure: the carrier remains rapid, but the packet envelope becomes finite. "Bandwidth smearing in optical interferometry: Analytic model of the transition to the double fringe packet" (Lachaume et al., 2012) writes the coherent flux as a carrier at central wavenumber $$f \approx \frac{\nu}{c}(\boldsymbol{b}(t)\times \omega_\oplus)\cdot \boldsymbol{n}$$9 multiplied by a packet envelope set by the spectral transmission. The packet width is about $\text{frequency}=\frac{W}{\lambda}$0, packet separation is $\text{frequency}=\frac{W}{\lambda}$1, and smearing becomes significant for

$\text{frequency}=\frac{W}{\lambda}$2

In the large-separation limit, one reaches the double or multiple fringe-packet regime (Lachaume et al., 2012). The important distinction is between carrier frequency in OPD space and envelope width in coherence-length space.

Fringe tracking instruments formalize the temporal counterpart. "The PRIMA fringe sensor unit" (0909.1470) uses spatial phase modulation and a low-resolution spectrometer to estimate phase and group delay at sampling rates up to $\text{frequency}=\frac{W}{\lambda}$3. "The GRAVITY fringe tracker" (Lacour et al., 2019) runs its detector at $\text{frequency}=\frac{W}{\lambda}$4, $\text{frequency}=\frac{W}{\lambda}$5, or $\text{frequency}=\frac{W}{\lambda}$6 Hz, uses phase delay for fast wrapped-phase control and group delay for slower packet tracking, and reports group-delay loop cutoffs of $\text{frequency}=\frac{W}{\lambda}$7, $\text{frequency}=\frac{W}{\lambda}$8, and $\text{frequency}=\frac{W}{\lambda}$9 Hz at those three sampling rates. Its practical correction bandwidth is about $I_n(\Phi)=|\mathcal{E}J_n(\Phi)|^2$0 Hz on the UTs and $I_n(\Phi)=|\mathcal{E}J_n(\Phi)|^2$1 Hz on the ATs, with OPD residuals as low as $I_n(\Phi)=|\mathcal{E}J_n(\Phi)|^2$2 nm RMS on the ATs and around $I_n(\Phi)=|\mathcal{E}J_n(\Phi)|^2$3 nm on the UTs because of structural vibrations (Lacour et al., 2019). In these systems, “fringe frequency” is most naturally interpreted as the temporal spectrum of OPD disturbances rather than a single carrier number.

Radio interferometry adopts yet another formalization: fringe rate as the Fourier pair of LST. "A demonstration of the effect of fringe-rate filtering in the Hydrogen Epoch of Reionization Array delay power spectrum pipeline" (Garsden et al., 2024) and "Optimized Beam Sculpting with Generalized Fringe-Rate Filters" (Parsons et al., 2015) use the approximate mapping

$I_n(\Phi)=|\mathcal{E}J_n(\Phi)|^2$4

so sky position, baseline geometry, observing frequency, and Earth rotation determine the temporal fringe rate of a source. HERA implements mainlobe and notch filters in fringe-rate space using discrete prolate spheroidal sequences, and the earlier beam-sculpting formalism shows that weighting fringe-rate modes is equivalent to applying a one-dimensional baseline-dependent weighting on the sky (Garsden et al., 2024, Parsons et al., 2015). This is an explicitly Fourier-dual usage of fringe frequency: a temporal mode coordinate used for sky selection, systematics rejection, and noise control.

6. Spectral and waveform fringes beyond classical interferometry

Several papers use “fringe frequency” for spectral oscillations rather than image-plane or time-domain carriers. "Alloharmonics in Burst Intensification by Singularity Emitting Radiation" (Ogura et al., 30 Jun 2025) studies BISER spectra with fine oscillations down to about $I_n(\Phi)=|\mathcal{E}J_n(\Phi)|^2$5, much smaller than the initial laser photon energy $I_n(\Phi)=|\mathcal{E}J_n(\Phi)|^2$6. The paper attributes this to laser frequency downshift in plasma and alloharmonic interference between different harmonic orders emitted at different times, with defining condition

$I_n(\Phi)=|\mathcal{E}J_n(\Phi)|^2$7

The fringe spacing follows

$I_n(\Phi)=|\mathcal{E}J_n(\Phi)|^2$8

and one fitted shot gave $I_n(\Phi)=|\mathcal{E}J_n(\Phi)|^2$9 (Ogura et al., 30 Jun 2025). In this usage, fringe frequency is literally the inverse spacing of spectral oscillations in photon-energy space.

"Probing Cosmic Strings with Gravitational-Wave Fringe" (Jung et al., 2018) uses the term for oscillations in detector frequency $f_{\rm width}$0 produced by coherent interference of lensed gravitational-wave paths. The closest formal quantity is the fringe width

$f_{\rm width}$1

with $f_{\rm width}$2. The cumulative number of oscillations scales as $f_{\rm width}$3, which explains why LIGO-band and mid-band detectors probe different tension ranges (Jung et al., 2018). The observable is not a spatial fringe but a regular modulation in the frequency-domain waveform.

A still different spectral use appears in "Confining Burst Energy Function and Spectral Fringe Pattern of FRB 20121102A with Multifrequency Observations" (Lyu et al., 2023). There the fringe pattern is a modulated probability distribution of burst spectral peak frequency $f_{\rm width}$4, modeled as a sum of Gaussian components over $f_{\rm width}$5–$f_{\rm width}$6 GHz. The inferred preferred $f_{\rm width}$7 values are $f_{\rm width}$8, $f_{\rm width}$9, $\omega$0, $\omega$1, $\omega$2, $\omega$3, $\omega$4, $\omega$5, and $\omega$6 GHz, with spacing of about $\omega$7 GHz. Combined with a single power-law burst-energy function,

$\omega$8

the simulations yield

$\omega$9

at $\text{frequency}=W/\lambda$00, and interpret the FAST bimodal energy distribution as a selection effect created by narrow spectra and a discrete $\text{frequency}=W/\lambda$01 pattern (Lyu et al., 2023). The paper explicitly notes that the putative $\text{frequency}=W/\lambda$02 fringe pattern cannot be explained with current radiation physics models (Lyu et al., 2023).

Taken together, these spectral and waveform studies extend fringe frequency far beyond classical interferograms. The unifying structure is interference among multiple emission times, paths, or preferred spectral loci, so that the measured fringe is an oscillation in photon energy, detector frequency, or event-population frequency rather than in image coordinates or OPD.

Across these literatures, fringe frequency functions less as a single definition than as a mathematically recurring role: it is the coordinate in which oscillatory interference becomes measurable and informative. Depending on the experiment, that coordinate may be OPD, sky Fourier space, LST-Fourier space, projector pixels, sideband order, detector frequency, or photon energy. The technical content of the term therefore lies in the measurement model that makes the fringe observable, not in a universal choice of units.