Adaptive phase locking in coherent beam combining with deep reinforcement learning.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Control Systems Engineering

Background:

• Locking of optical coherence by single-detector electronic-frequency tagging (LOCSET) is crucial for phase locking in coherent beam combining (CBC) systems.
• Current LOCSET methods require manual synchronization and hyperparameter tuning, limiting performance and adaptability to disturbances.
• The full potential of LOCSET remains untapped due to these manual limitations.

Purpose of the Study:

• To develop an adaptive phase-locking strategy for LOCSET in CBC systems using deep reinforcement learning.
• To enable automatic synchronization and dynamic optimization of LOCSET parameters in response to system disturbances.
• To overcome the limitations of manual control in existing LOCSET implementations.

Main Methods:

• Proposed a deep reinforcement learning-enabled adaptive phase-locking strategy comprising a Delay-Agent and an Ada-Agent.
• The Delay-Agent automatically synchronizes coherent demodulation by identifying optimal phase delay configurations.
• The Ada-Agent dynamically adjusts proportional coefficients and integration times to counter disturbances.

Main Results:

• Experimental validation in a four-channel CBC system demonstrated the effectiveness of the proposed strategy.
• The Delay-Agent successfully located the optimal phase delay combination in a single step.
• The Ada-Agent achieved a 4.5-fold improvement in recovery speed compared to standard LOCSET under various phase disturbances.

Conclusions:

• The proposed deep reinforcement learning-based adaptive phase-locking strategy significantly enhances LOCSET performance in CBC systems.
• This approach offers automatic synchronization and dynamic adaptability, overcoming limitations of manual control.
• The strategy holds potential for in situ application to improve phase-locking in diverse CBC systems.