TY - JOUR

T1 - Deep Reinforcement Learning Control of Mode-Locked Fiber Laser

AU - Serebrennikov, Kirill

AU - Kuprikov, Evgeny

AU - Kokhanovskiy, Alexey

N1 - The work of Alexey Kokhanovskiy was financially supported by ITMO Fellowship Program. The work of Kirill Serebrennikov was supported by the State budget of the Russian Federation (IA&E SB RAS project No FWNG2024-0015).

PY - 2025/12/8

Y1 - 2025/12/8

N2 - Fiber lasers are widely used source of pulsed radiation. One of the popular fiber laser schemes is laser based on the nonlinear polarization evolution (NPE) effect. Easy realization and ability to generate in different regimes are among advantages of such scheme. However, it demonstrates low level of robustness against the changes in external conditions and a nontrivial process of laser tuning. In this paper, we propose a novel approach to fiber laser tuning that is based on a deep reinforcement learning algorithm soft actor-critic. In this work, we demonstrated the algorithm's ability to extend the operating temperature up to certain limit without additional training sessions. This achievement highlights the algorithm's ability to recognize the fundamental behavioral patterns inherent in the laser system, rather than relying on rote memorization of context-dependent actions. Looking ahead, these findings suggest promising applications in real-time adaptive laser systems. By embedding self-correction mechanisms that respond to fluctuating environmental factors, such as temperature or mechanical stress, the methodology could revolutionize fields requiring high-stability laser operation, including telecommunications, precision metrology, and biomedical imaging.

AB - Fiber lasers are widely used source of pulsed radiation. One of the popular fiber laser schemes is laser based on the nonlinear polarization evolution (NPE) effect. Easy realization and ability to generate in different regimes are among advantages of such scheme. However, it demonstrates low level of robustness against the changes in external conditions and a nontrivial process of laser tuning. In this paper, we propose a novel approach to fiber laser tuning that is based on a deep reinforcement learning algorithm soft actor-critic. In this work, we demonstrated the algorithm's ability to extend the operating temperature up to certain limit without additional training sessions. This achievement highlights the algorithm's ability to recognize the fundamental behavioral patterns inherent in the laser system, rather than relying on rote memorization of context-dependent actions. Looking ahead, these findings suggest promising applications in real-time adaptive laser systems. By embedding self-correction mechanisms that respond to fluctuating environmental factors, such as temperature or mechanical stress, the methodology could revolutionize fields requiring high-stability laser operation, including telecommunications, precision metrology, and biomedical imaging.

KW - fiber lasers

KW - nonlinear polarization evolution

KW - reinforcement learning

KW - soft actor-critic

UR - https://www.scopus.com/pages/publications/105024572831

UR - https://www.mendeley.com/catalogue/d4fc8fb0-fe5b-32a9-b422-836086f5f291/

U2 - 10.1109/JLT.2025.3641711

DO - 10.1109/JLT.2025.3641711

M3 - Article

SP - 1

EP - 7

JO - Journal of Lightwave Technology

JF - Journal of Lightwave Technology

SN - 0733-8724

ER -