We review the use of machine learning techniques in ultrafast dynamics in fiber-optics systems. We discuss how neural networks can be used to correlate the spectral and temporal characteristics of dissipative soliton lasers and predict nonlinear dynamics in optical fibers for a wide range of input conditions. We also show how machine learning algorithm allow for optimizing supercontinuum generation.
We report for the first time the generation of a two-octave spanning supercontinuum (SC) from 700 nm to 2800 nm in a 20 cm non-silica graded-index multimode fiber. We study the SC generation and associated nonlinear instabilities in different dispersion regimes and characterize the SC stability. Significantly, under particular injection conditions, we observe clear signatures of self-cleaning dynamics with a near single-mode spatial intensity distribution at the fiber output. Our results are confirmed by numerical simulations of the 3D+1 generalized nonlinear Schrodinger equation
We use machine learning methods to control the spectral broadening experienced by femtosecond pulses in a highly nonlinear fiber. Combining a programmable spectral filter with a genetic algorithm or neural network allows us to optimize the nonlinear propagation dynamics to generate an on-demand target spectrum. Our approach is generic and can be adapted to a wide range of optical fibers and pump pulses. We expect our results to provide significant advances for adaptative control and tailored light sources.
The generation of an optical supercontinuum in nonlinear fibers exhibits highly complex nonlinear dynamics. Here, we show that one can train a neural network to learn the complex propagation dynamics for supercontinuum generation solely based on the input pulse parameters for a variety of scenarios ranging from higher-order soliton compression to broadband octave-spanning supercontinuum. The speed of our approach exceeds that of the direct integration of the generalized nonlinear Schrödinger equation by several orders of magnitude, allowing for “real-time” optimization or analysis of optical systems.
We show how neural networks can be used to model complex and predict nonlinear propagation dynamics in optical fibres for a widerange of input conditions and fibre systems, including pulse compression, ultra-broadband supercontinuum generation, and multimode fiber systems. Our results open up novel perspectives to model and optimize complex nonlinear dynamics and systems.
The generation of an optical supercontinuum with short (fs) input pulse duration is a highly complex process that exhibits rich nonlinear dynamics. Here, we show that one can teach a machine learning model to learn the nonlinear dynamics of ultrashort pulse propagation and predict the full-field propagation dynamics of supercontinuum based only on the input pulse characteristics (peak power, duration and chirp).
Although the successes of artificial intelligence in areas such as automatic translation are well known, the application of the powerful techniques of deep learning to current optics research is at a comparatively early stage. However, an area with particular promise for deep learning to accelerate both basic science and applications is in ultrafast optics, where nonlinear light-matter interactions lead to highly complex dynamics, including the emergence of extreme events. In the particular field of nonlinear fibre optics, we have recently reported a number of results that have shown how deep learning can both augment existing experimental techniques as well as provide new theoretical insights into the underlying physics. The objective of this paper is to review a selection of our work in this area.
Supercontinuum generation in the long pulse regime exhibits large shot-to-shot spectral variation and chaotic time domain consisting of soliton peaks emerging with random statistics. Under particular conditions, the noise-seeded dynamics may lead to the generation of a small number of extreme red-shifted rogue solitons that are associated with highly skewed “rogue wave” statistics. To overcome the restrictions in the experimental measurements, we here use the techniques of machine learning to predict the peak power and temporal shift of extreme red-shifted rogue solitons from single-shot spectral intensity profiles of supercontinuum without any phase information. The possibility to combine machine learning approaches with real-time spectral measurements to obtain temporal characteristics information without direct time-domain measurements which are often complex and limited to specific regimes of operations offers completely new avenues for the study of ultrafast dynamics in general.
Supercontinuum light (SC) is a broadband source with unique properties as the result of cascaded nonlinear dynamics when intense light propagates in a nonlinear material [1]. Recently, the generation of broadband SC sources operating in the mid-infrared (MIR) has attracted significant interest due to a wide range of potential applications in spectroscopy [2], microscopy, molecular fingerprinting [3], environmental monitoring and LIDAR [4], just to name a few. Fibers made of non-silica soft glasses such as fluoride, tellurite and chalcogenide are good candidates for SC generation in the MIR due to their high intrinsic nonlinearity and wide transparency window in this wavelength range. Supercontinuum generation in the MIR is typically achieved by propagating femtosecond pulses into the anomalous dispersion regime of a single-mode soft glass fibers. This typically leads to SC sources which are limited in terms of average power due to the low damage threshold associated with soft glasses and susceptible to noise due to the anomalous pump regime. These inherent limitations can be detrimental for many applications such as e.g. optical coherence tomography (OCT), photoacoustic imaging [5], or spectroscopy where the performance critically depends on the power and noise properties of the light source. In this work, we demonstrate for the first time the generation of a low noise high power SC in the MIR by injecting 350 fs pulses from an optical parametric amplifier into the normal dispersion regime of a one meter-long multimode step-index chalcogenide fiber with 100 µm core diameter. We show the generation of SC spanning from 1700 nm to 4800 nm for a pump wavelength at 3500 nm (located in normal dispersion regime of the used fiber) and an input peak power of 570 kW. A systematic study of the SC intensity noise is performed as a function of different pump parameters and for different output wavelengths show that the initial fluctuations on the pump laser are at most amplified by a factor of 2 at the spectral edge of the SC. Although the output beam in this case is multimoded, it can still be used for many practical applications such as long-distance remote sensing for which high power and low noise are more essential than the actual beam profile quality. Our results open novel perspective for the generation of high power low-noise broadband light sources in the MIR.
We will review our recent work in real-time measurements of nonlinear instabilities including the use machine learning, as well as the observation of a range of instability processes in novel dissipative systems such as the soliton-similariton laser.
Modulation instability (MI) is one of the most nonlinear instability of the focusing nonlinear Schrödinger equation that describes how a low amplitude noise on an input signal can be exponentially amplified to create high-intensity localised structures. Recently, MI has attracted renewed attention due to its potential link with the development of extreme events or rogue waves, and many theoretical, numerical and experimental studies have been reported in various physical systems including optics and hydrodynamics. A particular difficulty associated with experimental studies of MI in optics is the need for real-time measurement techniques. In the time domain, the time-lens approach is complex and constrains the measurement bandwidth and power. In the spectral domain, the dispersive Fourier transform is simpler but only typically allows for low dynamic range measurements and does not provide information about the associated temporal properties.
We report on our recent work on the use of machine learning to predict from real-time spectral data statistics for the maximum intensity of the localised temporal peaks in a fibre-optic chaotic MI field, peaks which are preferentially associated with extreme events. We subsequently train a neural network to correlate the spectral and temporal properties using data from numerical simulations and we use this model to predict the temporal probability distribution based on near 60 dB dynamic range real-time spectral data from experiments. These results open novel perspectives in all systems exhibiting chaos and instability where direct time-domain observations are difficult.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.