论文部分内容阅读
分析了一种光强解耦合的分布式随机并行梯度下降算法,此算法借助近场的波前传感器得到性能指标来得到算法的更新参数,这种性能指标解耦了随机并行梯度下降算法使用的耦合的全场光强,使得算法性能得到提升。分析了一种马赫泽得形式的自参考点衍射干涉仪作为波前传感器。建立两种仿真模型对算法进行了分析,结果表明分布式的随机并行梯度下降算法比原算法在收敛速度上有了数量级的提升,在127仿真结果单元模型上,收敛速度是原算法的10倍以上且收敛结果几乎相同。仿真研究针对不同光束之间的平移相差和同光束的高阶像差,显示了算法应用在光束相干合成的前景。
A distributed random and parallel gradient descent algorithm with light intensity decoupling is analyzed. This algorithm obtains the updated parameters of the algorithm by using the near field wavefront sensor to get the performance index, which decouples the random parallel gradient descent algorithm Coupled full-field light intensity, making the algorithm performance is improved. A kind of Mach-Zehnder self-reference diffraction interferometer is analyzed as a wavefront sensor. Two kinds of simulation models are established to analyze the algorithm. The results show that the distributed stochastic gradient descent algorithm has an order of magnitude improvement in convergence speed compared with the original algorithm. On the simulation results unit model of 127, the convergence speed is 10 times that of the original algorithm And the convergence results are almost the same. The simulation study shows the application of the algorithm in the prospect of coherent beam synthesis in view of the phase difference between different beams and the higher-order aberrations of the same beam.