High-Linear Frequency-Swept Lasers with Data-Driven Control

Abstract

The frequency-swept laser (FSL) is applied widely in various sensing systems in the scientific and industrial fields, especially in the light detection and ranging (Lidar) area. However, the inherent nonlinearity limits its performance in application systems, especially in the broadband frequency-swept condition. In this work, from the perspective of data-driven control, we adopt the reinforcement learning-based broadband frequency-swept linearization method (RL-FSL) to optimize the control policy and generate the modulation signals. The nonlinearity measurement system and the system simulator are established. Since the powerful learning ability of the reinforcement learning algorithm, the linearization policy is optimized off-line and the generated modulation signals reduce the nonlinearity almost 20 times, compared to the case without control. In the long-term operation, the regular updated modulation signals perform better than the traditional iteration results, demonstrating the efficiency of the proposed data-driven control method in application systems. Therefore, the RL-FSL method has the potential to be the candidate of optical system control.

Keywords

• frequency-swept lasers
• reinforcement learning
• nonlinearity
• data-driven control