Reinforcement learning-driven spectral broadening in NPR laser: 114.3-nm bandwidth generation with sub-minute optimization

Closed-loop control of a nonlinear polarization rotation (NPR) mode-locked fiber laser is achieved using a deep reinforcement learning (DRL) algorithm integrated with a long short-term memory (LSTM) network to guide the spectral broadening strategy. This approach overcomes conventional tuning limitations in NPR systems. A record-high stable mode-locking bandwidth of 114.3 nm (maintained over 4 hours) and a pulse duration of 109.37 fs are achieved. Vibration resistance tests show that the trained model recovers stable mode-locking and reaches the maximum spectral width in an average time of 4.06 s. This automatic control and adaptive spectral broadening highlight the potential of advanced ultrafast laser sources in spectral sensing and precision metrology.