涡旋光束经过大气湍流时,其波前会发生畸变,因此需要对畸变的波前进行校正。无波前传感器的波前畸变校正系统基于随机并行梯度下降算法,可以实现对波前畸变的校正,但算法的收敛速度及稳定性受随机扰动电压的影响。结合深度学习理论中改进的梯度下降算法,对随机并行梯度下降算法中随机扰动电压的迭代方式进行调整,并分析不同湍流强度下改进型算法的校正效果。仿真结果表明:在弱湍流条件下,需优先选择基于RMSprop的改进型算法;而在中等湍流和强湍流条件下则需要结合实际需求从算法的稳定性、性能评价函数大小以及收敛速度等方面考虑,选择合适的校正算法。

When the vortex beam passes through atmospheric turbulence, the wavefront will be distorted, and the distorted wavefront needs to be corrected. The wavefront distortion correction system of wavefrontless sensors based on the stochastic parallel gradient descent algorithm can realize the correction of wavefront distortion, but the convergence speed and stability of the algorithm are affected by the random disturbance voltage. Combined with the improved gradient descent algorithm in deep learning theory, the iteration method of the random disturbance voltage in the stochastic parallel gradient descent algorithm is adjusted in this paper, and the correction effect of the improved algorithm under different turbulence intensities is analyzed. The simulation results show that the improved algorithm based on RMSprop is preferred under weak turbulence condition. Under moderate turbulence and strong turbulence conditions, it is necessary to consider the stability of the algorithm, the value of the performance evaluation function, and the convergence speed in accordance with actual requirements, and then select the appropriate correction algorithm.