分析随机并行梯度下降(SPGD)算法用于多路大型固体激光装置相干合成中校正动态相差的能力。首先介绍了SPGD算法实现相干合成的基本理论,利用数值模拟方法对算法进行了优化,实现了两路基于SPGD算法的波长为800 nm、带宽为30 fs光束的相干合成实验,验证了在外加10,15,20,25 Hz动态相差条件下算法的特性,并进一步模拟了动态活塞相差和指向性相差的校正过程,分析了不同相位噪声强度和频率对校正能力的影响,计算了控制带宽与光束路数、算法执行速度之间的关系。结果表明:远场强度分布的平方和是高能短脉冲激光相干合成的最佳性能评价函数;采用自适应增益的方式时,在保证算法稳定性的前提下,提高了算法的收敛速度;随着相位噪声强度和频率的提高,算法的有效控制带宽减小;算法执行速度越快,光束路数越少,则算法控制带宽越大;受限于器件性能,SPGD算法不适用于4路以上带宽为30 fs激光阵列的相干合成。

The correction ability of dynamic phase error in coherent beam combination for multi-channel large solid laser device is analyzed by stochastic parallel gradient descent (SPGD) algorithm. The fundamental theory of coherent beam combination by SPGD algorithm is introduced. The algorithm is optimized by numerical simulation method. Coherent beam combination of two beams with the wavelength of 800 nm and the bandwidth of 30 fs is experimentally achieved and the performances of SPGD algorithm under the 10, 15, 20, 25 Hz dynamic phase error conditions are tested. Moreover, the processes of dynamic piston and point phase error correction are simulated. The influence of phase noises with different intensities and frequencies on the correction ability is analyzed. The relationships among the control bandwidth, number of beams and iteration rate are computed. The results show that the quadratic sum of far-field intensity is the optimal performance evaluation function of coherent beam combination for high-power short-pulse laser. The adaptive gain can guarantee the stability and improve the convergence speed of the algorithm. The effective control bandwidth decreases with increasing intensity or frequency of phase noise, and increases with increasing iteration rate and decreasing of beam number. Limited by the performance of device,SPGD algorithm cannot be applied to the coherent beam combination for more than four beams laser array with the bandwidth of 30 fs.