Analysis of phase instabilities in amplitude swing for ultrashort pulse train characterization

The temporal dynamics of ultrashort pulses are a fundamental feature in ultrafast optics. These dynamics can often be extracted from a two-dimensional trace consisting of a set of nonlinear spectra, using an iterative algorithm. Typically, the measurement of this trace requires integrating the signal of many pulses, which implies that the trace does not correspond to a single pulse when shot-to-shot variations occur. In this case, the pulse train can be characterized statistically, by a base pulse and a metric that quantifies the instabilities. Here, we demonstrate that the amplitude swing technique is sensitive to instabilities, which manifest as two distinct imprints on the measurement. First, we analyze the terms that compose the amplitude swing signal. Then, we study two parameters to quantify the instabilities, related to a shift in the minima of the trace marginal, and to the filling of the trace minima zones, respectively. Finally, we apply these strategies to characterize simulated unstable pulse trains, using a simple technique.

💡 Research Summary

The paper investigates how the amplitude‑swing (a‑swing) technique, a variant of nonlinear‑spectral‑trace methods, responds to shot‑to‑shot phase fluctuations in an ultrashort‑pulse train. Conventional 2‑D trace techniques such as FROG, d‑scan, and FROSt assume that each pulse in the train is identical, so the measured trace—obtained by integrating many pulses—represents a single pulse. When this assumption fails, the trace becomes an incoherent superposition of many individual traces and standard retrieval algorithms no longer yield a unique solution. The authors therefore adopt a statistical approach: they model the train as a “base” pulse plus a metric that quantifies the variability.

First, they present a detailed Fourier‑based analysis of the a‑swing trace. In the a‑swing setup, two time‑delayed replicas of the unknown pulse are combined using a multiple‑order waveplate (MWP) and a linear polarizer (LP). The relative amplitude between the replicas is encoded by the MWP rotation angle θ. After second‑harmonic generation (SHG) in a nonlinear crystal, the intensity I(ω,θ)=|E(ω,θ)|² is recorded as a function of optical frequency ω and angle θ. By expanding the electric field in cos θ and sin θ components, the intensity is shown to consist only of even‑order cosine terms up to the eighth order. The coefficients sₙ(ω) of these terms are directly proportional to the Fourier coefficients of the trace, and the marginal distributions M_ω(ω) and M_θ(θ) are simple linear combinations of s₀, s₂, s₄, etc. Consequently, the trace alone contains enough information to infer the sign of the spectral phase and to detect variations without any prior knowledge of the pulse.

To test sensitivity to instabilities, the authors generate five simulated pulse trains. Each train consists of 101 pulses derived from a base pulse with a Gaussian spectrum centered at 800 nm, a Fourier‑limited duration of 50 fs, and a fixed spectral phase (‑3000 fs² GDD + 10⁵ fs³ TOD). Random phase perturbations are added, controlled by a parameter ε (0.15, 0.30, 0.45, 0.60, 0.75). The random phase is smoothed with a third‑order super‑Gaussian temporal filter (1200 fs width) to emulate realistic low‑frequency phase noise. As ε increases, both the full‑width‑at‑half‑maximum (FWHM) and the second‑moment (SM) pulse durations grow, indicating broader, more structured pulses.

The a‑swing traces are then simulated for two MWP phase retardations: 1.5π (quarter‑wave‑plate operation) and π (half‑wave‑plate operation). Retrieval is performed with the ptychographic algorithm previously developed for a‑swing. Two distinct signatures of instability emerge:

1. MWP retardation = 1.5π – The Fourier coefficients s₂(ω) and s₆(ω) change sign with increasing ε, causing a shift of the minima in the angular marginal M_θ(θ). The authors define a metric η_shift = Δθ_min / 90°, where Δθ_min is the displacement of the marginal minimum. η_shift is zero for a perfectly stable train and approaches one as ε grows, providing a clear quantitative measure of phase noise.

2. MWP retardation = π – The trace’s low‑intensity zones (“null zones”) become partially filled as ε increases. The standard G‑error (the normalized root‑mean‑square difference between simulated and retrieved traces) remains low (<2 %) because the discrepancy is confined to these null zones, making G‑error an unreliable instability indicator. Instead, the authors compute the normalized mean‑square error within the whole trace, η_fill = ⟨E²⟩ / ⟨I_sim²⟩, where E(ω,θ) is the pointwise difference between simulated and retrieved intensities. η_fill also ranges from 0 (stable) to 1 (highly unstable) and captures the “filling” effect.

Both metrics can be extracted directly from the measured trace without any iterative retrieval, making them attractive for real‑time monitoring. The paper also notes that the Fourier‑coefficient analysis can be extended to other a‑swing configurations (different MWP orders, alternative nonlinear media), potentially enabling more sophisticated retrieval strategies that weight specific coefficients.

In the discussion, the authors emphasize that while the a‑swing technique has previously been valued for its simplicity and robustness under stable conditions, this work reveals its intrinsic sensitivity to statistical pulse‑to‑pulse variations. The proposed metrics overcome the limitation of conventional error measures and provide a practical pathway to diagnose and possibly correct instabilities in high‑repetition‑rate laser systems where single‑shot measurements are impractical.

Overall, the study demonstrates that amplitude‑swing measurements can serve not only as a pulse‑characterization tool but also as a diagnostic for phase instabilities, offering a low‑complexity, statistically‑based alternative to more elaborate single‑shot techniques.