Towards ‘smart lasers’: self-optimisation of an ultrafast pulse source using a genetic algorithm

Abstract

Short-pulse fibre lasers are a complex dynamical system possessing a broad space of operating states that can be accessed through control of cavity parameters. Determination of target regimes is a multi-parameter global optimisation problem. Here, we report the implementation of a genetic algorithm to intelligently locate optimum parameters for stable single-pulse mode- locking in a Figure-8 fibre laser, and fully automate the system turn-on procedure. Stable ultrashort pulses are repeatably achieved by employing a compound fitness function that monitors both temporal and spectral output properties of the laser. Our method of encoding photonics expertise into an algorithm and applying machine-learning principles paves the way to self-optimising ‘smart’ optical technologies.

Iteratively seeded mode-locking

“…steps is not be crucial. Methods already exist for tuning resonator components such as the energy, dispersion, saturable absorber, and spectral filter [19][20][21][22][23][24][25][26]. A recent demonstration of control over mode-locking states with a programmable spatial light modulator inside of a laser resonator [27] demonstrates that robust experimental architectures exist and have been developed which are capable of implementing ISM designs.…”

“…multi-pulsing instabilities [14,38]) and select desired pulse solutions within a complicated resonator phase space of competing pulse solutions, demonstrates the importance of designing a resonator to support pulse formation within a desired basin of attraction. Research examining how to improve pulse states through programmable wave plates, demonstrates the effect that relatively small one-dimensional changes to cavity elements can have in the local optimization of a pulse quality [19][20][21][22][23][24][25][26]. Recent research demonstrating reversible and irreversible pulse state transitions through use of an SLM inside of a resonator [27], and observation of hysteresis effects [39], point to the importance which the path taken to a potential solution can have in stabilization of that solution.…”

“…Specifically, researchers have implemented algorithmic and machine learning approaches to resonator parameter control, which, in concert with a suitable figure of merit, can help optimize a resonator for a certain pulse quality such as a minimum pulse width or a high peak power [19][20][21][22][23][24][25][26][27].…”

Investigation of ellipticity and pump power in a passively mode-locked fiber laser using the nonlinear polarization rotation technique

“…Different PC settings yield different spectrum energies, durations, shapes, and widths. Oliver et al [19][20][21] demonstrated this by measuring the output polarization state, analyzing the inter-mode beat spectrum and by determining the allowed operating regimes. In this work, the relationship between pump power levels and the ellipticity of a mode-locked fiber laser of the NPR technique is successfully demonstrated.…”

Generation of 70-fs pulses at 286 μm from a mid-infrared fiber laser

“…By careful adjustment of the waveplates, the cavity is biased to preferentially operate in a mode-locked state, suppressing CW emission. Once the correct waveplate angles are identified (which in future will be algorithmically automated [16,17]), mode-locking is reliably self-starting at ∼3 W pump power. A stable pulse train is generated with 54 MHz repetition rate and up to 200 mW average power (3.7 nJ pulse energy).…”