Intelligent Ultrafast Photonics Based on Machine Learning：Review and Prospect（Invited）

Jiajun PENG; Xiaohui LI; Sunfan XI; Keqin JIAO

doi:10.3788/gzxb20225108.0851518

Journals >Acta Photonica Sinica >Volume 51 >Issue 8 >Page 0851518 > Article

Acta Photonica Sinica
Vol. 51, Issue 8, 0851518 (2022)

Intelligent Ultrafast Photonics Based on Machine Learning：Review and Prospect（Invited）

Jiajun PENG^1、1, Xiaohui LI^1、1, Sunfan XI^1、1, and Keqin JIAO^1、1

Author Affiliations

¹1School of Physics and Information Technology，Shaanxi Normal University，Xi'an 710119，China

¹2College of Life Sciences，Shaanxi Normal University，Xi'an 710119，China

show less

DOI: 10.3788/gzxb20225108.0851518 Cite this Article

Jiajun PENG, Xiaohui LI, Sunfan XI, Keqin JIAO. Intelligent Ultrafast Photonics Based on Machine Learning：Review and Prospect（Invited）[J]. Acta Photonica Sinica, 2022, 51(8): 0851518 Copy Citation Text

show less

Infrastructure of machine-learning strategies for automatic mode-locking ultrafast fibre lasers using control of intracavity elements via a feedback loop and control algorithm

Fig. 1. Infrastructure of machine-learning strategies for automatic mode-locking ultrafast fibre lasers using control of intracavity elements via a feedback loop and control algorithm

Download full size | View in the Article

Fig. 2. Algorithm policies

Download full size | View in the Article

Fig. 3. The main principles and design of ultrafast photonics under machine learning strategies

Download full size | View in the Article

Laser system	Control elements	Objective function	Algorithms	Targeted parameters	Advantages	Disadvantages	Performance	References
Mode-locked fibre laser	Waveplates，polarizer，birefringence	Pulse energy divided by spectral kurtosis of the waveform	Search algorithm，singular value decomposition，local optimization algorithm	Stable mode locking	Fast identification of birefringence state and optimization of controller parameters	Requires a large number of measured parameters	10 min~1 h	［16，61，120，121］
NPE fibre laser	Electronic polarization controller，liquid crystal variable retarder	FFT，corresponding function of different systems，RF peak	Advanced search algorithm，random collision recovery，genetic algorithm，human-like algorithm	Fundamental and harmonic mode locking，Q-switching and mode locking	Multi-function，real-time，multi-regimes of operation	Instability cannot be detected in real time	Mode-locking time of a few seconds，subsecond recovery time	［62，63，105，128］
NPE fibre laser	Electronic polarization controller	Anomalous Harmonic power for anomalous dispersion，intensity of FSR radio-frequency component for normal dispersion	Evolutionary algorithm	Q-switching and stable mode locking	Two regimes of operation	Slow convergence	Mode-locking time of 30 min	［54］
NPE fibre laser	Polarization control	Repetition frequency	Evolutionary algorithm	Harmonic mode-locking with anomalous dispersion	Optimized high harmonic mode locking	Slow convergence	Harmonic mode-locking time of 2 h	［122］
Ring fibre laser	Electronic polarization controller，pump power	Peak power，maximized RF signal	Genetic algorithm	Anomalous dispersion with stable single-pulse mode locking	High contrast between stable and unstable pulses	Complex fitness function，slow convergence speed	Mode-locking time of 30 min	［57］
Ring fibre laser	Electronic polarization controller，electric dial，Normal dispersion	Centre wavelength，coefficient between pulse amplitude jitter spectrum and target	Genetic algorithm	Stable tunable，birefringent filtering	Adjust center wavelength and repetition frequency	Limited tuning	Unable to integrate	［58，123，［126］
NPE fibre laser	Electronic polarization controller	Radio-frequency power at expected repetition rate，spectral similarity and output power	Genetic algorithm	Stable mode locking	Spectra can be tuned	Only fundamental mode locking	Mode-locking time of 90 seconds，30 s recovery time	［124］
NPE fibre laser	Polarizer，amplifier，counter	Pulse energy of single pulse solution	Genetic algorithm	Multipulse mode	Simple fitness function	Complex polarization control	Numerical simulation results	［55，125］
Figure-of-eight laser	Pump diode powers	Autocorrelation duration	XGBoost，Feed-forward neural network	Replace time domain comb	Real-time multiparameter monitoring with a single oscilloscope	Requires a large number of measured parameters	Unable to integrate practicality	［104］
Mode-locked fibre laser	Waveplates，polarizer	Pulse energy divided by spectral kurtosis of the waveform	Feed-forward neural network，recurrent neural network	Stable mode locking	Fiber birefringence changes quickly	Slow training process	Numerical simulation results	［76］

Table 1. Comparison of design methods and performance of ultrafast fiber lasers based on machine learning strategy

Jiajun PENG, Xiaohui LI, Sunfan XI, Keqin JIAO. Intelligent Ultrafast Photonics Based on Machine Learning：Review and Prospect（Invited）[J]. Acta Photonica Sinica, 2022, 51(8): 0851518

Download Citation

Tools

Save the article for my favorites

Paper Information