TY - JOUR

T1 - Reinforcement Learning for Stabilization of Self-Injection Locking in a DFB Laser with an External Fiber Ring Cavity

AU - Gemuzov, A. S.

AU - Itrin, P. A.

AU - Panyaev, I. S.

AU - Korobko, D. A.

AU - Redyuk, A. A.

AU - Fotiadi, A. A.

AU - Fedoruk, M. P.

AU - Bednyakova, A. E.

N1 - A. S. Gemuzov et al., "Reinforcement Learning for Stabilization of Self-Injection Locking in a DFB Laser With an External Fiber Ring Cavity," in IEEE Journal of Selected Topics in Quantum Electronics, vol. 32, no. 5: Self-Injection Locked Lasers and Assoc. Sys., pp. 0900511-0900511, Sept.-Oct. 2026, Art no. 0900511, doi: 10.1109/JSTQE.2026.3712262. This work was supported by the Russian Science Foundation under Project 25-61-00010, https://rscf.ru/project/25-61-00010/ (the development and implementation of the neural-network control system) and Project 23-79-30017 (the design and testing of the SIL fiber laser). (Corresponding authors: Anastasia E. Bednyakova; Andrei A. Fotiadi.)

PY - 2026/9

Y1 - 2026/9

N2 - Self-injection locking of semiconductor lasers to high-Q resonators provides an effective route to linewidth narrowing, low phase noise, and high frequency stability. Here, we demonstrate reinforcement-learning-based stabilization of a self-injection-locked distributed-feedback (DFB) laser using an external polarization-maintaining fiber ring cavity assembled from standard telecommunication components. Narrow-linewidth operation is achieved through coherent optical feedback from the cavity resonance, while long-term retention of the locked state is supported by a simple active optoelectronic loop incorporating a thermo-optic phase shifter. The stabilization task is formulated as a continuous-control reinforcement learning problem in which a neural-network controller regulates the optical phase in the feedback path to maintain the photodetector signal at a desired reference level. To enable real-time learning on the physical setup, an asynchronous architecture is implemented that decouples data acquisition from policy optimization. Compared with conventional PID control, the learned controller provides more robust locking retention, avoids abrupt phase-reset-driven disruptions, and yields a narrower error distribution centered closer to the target level. These results establish reinforcement learning as a practical tool for intelligent stabilization of compact narrow-linewidth self-injection-locked laser sources.

AB - Self-injection locking of semiconductor lasers to high-Q resonators provides an effective route to linewidth narrowing, low phase noise, and high frequency stability. Here, we demonstrate reinforcement-learning-based stabilization of a self-injection-locked distributed-feedback (DFB) laser using an external polarization-maintaining fiber ring cavity assembled from standard telecommunication components. Narrow-linewidth operation is achieved through coherent optical feedback from the cavity resonance, while long-term retention of the locked state is supported by a simple active optoelectronic loop incorporating a thermo-optic phase shifter. The stabilization task is formulated as a continuous-control reinforcement learning problem in which a neural-network controller regulates the optical phase in the feedback path to maintain the photodetector signal at a desired reference level. To enable real-time learning on the physical setup, an asynchronous architecture is implemented that decouples data acquisition from policy optimization. Compared with conventional PID control, the learned controller provides more robust locking retention, avoids abrupt phase-reset-driven disruptions, and yields a narrower error distribution centered closer to the target level. These results establish reinforcement learning as a practical tool for intelligent stabilization of compact narrow-linewidth self-injection-locked laser sources.

KW - Self-injection locking

KW - machine learning methods

KW - narrow-band lasers

KW - reinforcement learning

KW - Самосинхронизация инжекцией

KW - узкополосные лазеры

KW - методы машинного обучения

KW - обучение с подкреплением

UR - https://www.mendeley.com/catalogue/c70c18eb-7fc0-3e24-87c2-49d2c11c4bd8/

UR - https://www.scopus.com/pages/publications/105044738140

U2 - 10.1109/JSTQE.2026.3712262

DO - 10.1109/JSTQE.2026.3712262

M3 - Article

VL - 32

SP - 900511

EP - 900511

JO - IEEE Journal of Selected Topics in Quantum Electronics

JF - IEEE Journal of Selected Topics in Quantum Electronics

SN - 1077-260X

IS - 5

ER -