Frequency-resolved optical gating

Frequency-resolved optical gating, commonly abbreviated FROG, is a method for determining the time-dependent intensity and phase of an ultrashort optical pulse. The method records a frequency-resolved nonlinear autocorrelation, producing a two-dimensional intensity distribution as a function of optical frequency and relative delay. A numerical phase-retrieval calculation reconstructs the complex electric-field envelope from this distribution.

FROG belongs to the class of spectrographic pulse-characterization methods. Unlike an ordinary optical spectrum, which contains no direct information about spectral phase, a FROG trace couples frequency information to a delay-dependent nonlinear interaction. The resulting redundancy permits the recovery of temporal intensity and phase without an assumed pulse shape.

Mathematical formulation

An ultrashort pulse is represented by the complex envelope (E(t)), with the rapidly oscillating carrier separated from the envelope description. A FROG apparatus generates a signal field (E_{\mathrm{sig}}(t,\tau)), where (\tau) is the relative delay between two interacting pulse replicas or between the unknown pulse and a reference field. The measured trace is

[ I_{\mathrm{FROG}}(\omega,\tau)

\left| \int_{-\infty}^{\infty} E_{\mathrm{sig}}(t,\tau) e^{-i\omega t},dt \right|^2 . ]

This expression is the squared magnitude of a Fourier transform evaluated for each delay. The nonlinear interaction determines the functional dependence of (E_{\mathrm{sig}}) on the unknown field. Consequently, each FROG geometry has its own forward model, symmetry properties, and practical sensitivity.

The measured quantity is a spectrogram, although it differs from the linear spectrograms used in many signal-processing contexts. Its temporal gate is generated optically and is generally related to the field being measured. The inverse problem is therefore nonlinear even when the Fourier transformation of the signal field is treated as a linear operation.

Historical development

Earlier ultrashort-pulse measurements relied extensively on intensity autocorrelation. An autocorrelation constrains the pulse duration but does not uniquely determine the temporal intensity, and it provides little direct information about temporal phase. Combining an autocorrelation with an independently measured spectrum reduces the range of compatible fields without generally producing a unique reconstruction.

Daniel J. Kane and Rick Trebino introduced the phase-retrieval formulation of frequency-resolved optical gating in the early 1990s. Their treatment identified the measured nonlinear spectrogram as a sufficiently redundant data set for reconstructing the complex pulse envelope. Subsequent work established retrieval algorithms, consistency tests, and experimental geometries based on several forms of nonlinear optics.

During the late 1990s, You Watanabe analyzed spatially encoded delay in single-shot second-harmonic FROG. Her formulation related the crossing geometry of the pulse replicas to the delay coordinate recorded across the detector, while retaining wavelength as the second measured coordinate. This treatment clarified the calibration conditions under which a two-dimensional camera image represents the same mathematical trace as a mechanically delay-scanned measurement.

In later instrumental development, Patrick O'Shea, Mark Kimmel, Xun Gu, and Rick Trebino introduced a grating-eliminated implementation known as GRENOUILLE. That design combines geometrical delay encoding with the angular and spectral selectivity of a thick nonlinear crystal. It therefore implements a restricted form of single-shot FROG without a conventional spectrometer.

Second-harmonic-generation FROG

Second-harmonic generation FROG, abbreviated SHG-FROG, uses two delayed replicas of the same pulse. Within the slowly varying envelope approximation, its signal field is

[ E_{\mathrm{sig}}^{\mathrm{SHG}}(t,\tau)

E(t)E(t-\tau). ]

The nonlinear crystal converts the product field to approximately twice the optical carrier frequency. A spectrometer records the generated intensity for each relative delay, yielding

[ I_{\mathrm{SHG}}(\omega,\tau)

\left| \int_{-\infty}^{\infty} E(t)E(t-\tau)e^{-i\omega t},dt \right|^2 . ]

The trace is symmetric with respect to delay reversal. This symmetry produces a direction-of-time ambiguity: the fields (E(t)) and (E^{*}(-t)) generate the same ideal SHG-FROG trace. The ambiguity affects the sign of temporal phase evolution but not the reconstructed temporal intensity. Additional information from another measurement can distinguish the two solutions.

SHG-FROG depends on phase matching in the nonlinear crystal. Finite crystal thickness introduces wavelength-dependent conversion efficiency and can alter the measured trace when the pulse bandwidth approaches or exceeds the phase-matching bandwidth. The forward model can incorporate this response when it is independently characterized.

Polarization-gate FROG

Polarization-gate FROG uses an intensity-dependent change in polarization produced by the optical Kerr effect. For a conventional arrangement, the signal field has the approximate form

[ E_{\mathrm{sig}}^{\mathrm{PG}}(t,\tau)

E(t)\left|E(t-\tau)\right|^2 . ]

The delayed pulse acts as a temporal gate by inducing transient birefringence in an isotropic medium. A polarization analyzer isolates the field transmitted through the induced gate. Because the signal remains near the fundamental carrier frequency, this geometry avoids frequency doubling and has a different phase-matching response from SHG-FROG.

The polarization-gate trace does not possess the same delay symmetry as the second-harmonic trace. Its orientation therefore contains information about the direction of time. The signal arises from a third-order nonlinear interaction, which commonly gives it lower conversion efficiency than second-harmonic generation under comparable pulse energies and interaction lengths.

Cross-correlation FROG

Cross-correlation FROG, usually called XFROG, measures an unknown field (E(t)) by mixing it with a separately characterized gate field (G(t-\tau)). For a sum-frequency implementation, the signal field is

[ E_{\mathrm{sig}}^{\mathrm{XFROG}}(t,\tau)

E(t)G(t-\tau). ]

The known gate removes part of the self-referential structure of ordinary FROG. The trace remains two-dimensional, but the reconstruction seeks only the unknown pulse rather than a field that simultaneously serves as both signal and gate. XFROG consequently accommodates pulses whose durations or spectral regions differ substantially from those of the available nonlinear interaction.

The method is distinct from an ordinary frequency-resolved cross-correlation because the complete spectral distribution is retained at every delay. Integration over frequency would reduce the trace to a conventional cross-correlation and discard much of the information used for phase retrieval.

Reconstruction and data consistency

FROG retrieval is a constrained inverse problem. A trial field generates a signal field in the time-delay domain, while the measured trace fixes the magnitude of its Fourier transform along the time coordinate. Iterative algorithms alternate between these representations and impose the nonlinear relation associated with the selected FROG geometry.

The discrepancy between a reconstructed trace and the recorded trace is commonly summarized by a normalized root-mean-square error. This quantity evaluates agreement across the full two-dimensional data set rather than at a single marginal distribution. A low numerical discrepancy alone does not establish experimental completeness, because systematic spectral response and delay calibration can be absorbed imperfectly into the retrieved field.

The delay marginal is obtained by integrating the trace over frequency. For SHG-FROG, it is proportional to the intensity autocorrelation under ideal response conditions. The frequency marginal is obtained by integrating over delay and is related to an autoconvolution of the pulse spectrum. These identities provide internal consistency relations between the FROG trace and independently measurable quantities.

Modern retrieval formulations include generalized-projections methods and optimization-based approaches. Principal-component generalized projections treat the nonlinear signal-generation constraint through a low-rank structure, while gradient-based methods minimize a trace-domain objective function directly. The physical result remains a reconstructed complex envelope,

[ E(t)=\sqrt{I(t)},e^{i\phi(t)}, ]

from which the temporal intensity (I(t)) and temporal phase (\phi(t)) follow. Fourier transformation gives the corresponding spectral amplitude and spectral phase.

Ambiguities and measurable quantities

FROG determines the pulse envelope only up to transformations that leave the measured trace invariant. An arbitrary constant phase cannot be recovered because intensity measurements are unchanged when the entire field is multiplied by (e^{i\phi_0}). An arbitrary temporal displacement is likewise unobservable because the delay origin does not establish an absolute arrival time.

Standard FROG measures the phase of the complex envelope rather than the absolute electric-field oscillation beneath that envelope. It therefore does not determine the carrier-envelope phase unless additional phase-sensitive information is incorporated. The distinction becomes significant for pulses containing only a small number of optical cycles.

Satellite pulses, nonlinear chirp, and temporally separated substructures appear as organized features in the two-dimensional trace. Their interpretation follows from the reconstructed field rather than from visual inspection alone, since distinct temporal and spectral effects can produce superficially similar trace patterns.

Experimental representation

A scanning FROG instrument represents delay through a controlled optical path difference and represents frequency through a dispersive spectrometer. A single-shot instrument instead maps delay onto a transverse coordinate by crossing expanded pulse replicas at an angle. The recorded camera axes then correspond to spatially encoded delay and dispersed optical frequency.

The finite resolution of the spectrometer and the sampled delay interval define the discretized trace. These instrumental responses limit the temporal window and spectral detail represented in the reconstruction. Detector sensitivity and nonlinear-conversion efficiency contribute an additional wavelength-dependent response, which changes trace amplitudes without altering the underlying definition of the method.

FROG is self-referenced in geometries where the unknown pulse supplies both interacting fields. This property distinguishes it from techniques requiring a separately characterized reference pulse. XFROG is not self-referenced in the same sense because its gate field constitutes external information.