Neural Network Assisted Laser Frequency Locking System

Abstract

The laser frequency locking system with high robustness, high accuracy, and good anti-noise performance is usually used in laser synchronization and optics communications, continuous-variable quantum key distribution, and ultra-stable laser wavelength stabilization. In such systems, the analog or digital method is usually used to implement feedback control. Digital systems are more widely used than analog systems based on phase-locked loops due to their simplicity, flexibility, and robustness. The proportional- integral-derivative (PID) algorithm is the most typical algorithm well-developed in the digital locking system. However, due to the nonlinearity of the error signal induced by frequency variation, the PID algorithm will have a deviation in compensation for the transient response. This paper demonstrates a neural network assisted laser frequency locking system using predicted values to decrease the nonlinearity. Compared to the PID algorithm locking performance in the same locking system, both the peak-to-peak deviation and the root- mean-square error of the neural network assisted locking system are reduced by half.