Physics Based Phase Control in Coherent Beam Combining Systems

In the realm of high-power lasers, coherent beam combining has become a scalable and dependable technology. However, achieving phase synchronization in a coherent beam-combining (CBC) system is a significant challenge. The predominant approach for tackling this challenge is the utilization of advanced optimization techniques such as the stochastic parallel gradient descent (SPGD) method. However, this method comes with notable drawbacks, including slow convergence, potential convergence to local extremes, and scalability limitations. In this paper, we introduce a new physics-based approach designed to effectively optimize the phases of individual laser elements within the CBC system. The key feature of our approach is the use of deep learning optimizers as against the traditional approach for computing the gradients using numerical differentiation like in the case of SPGD. To prove the effectiveness of the approach, we present a comprehensive comparative analysis incorporating a range of different deep-learning optimizers for a higher number of laser elements under different operational settings. The results revealed that the proposed algorithms consistently outperform the conventional methods across all aspects, even in a dynamic environment where the phase noise changes with time. Among the various algorithms tested, Adagrad has proven particularly robust and effective, requiring only a minimal number of iterations to converge. Our research, besides identifying a suitable algorithm for phase synchronization in CBC systems, also sheds light on the effectiveness of the deep learning algorithm for optimization tasks and, therefore, shall serve as a reference for multiparameter global optimization problems.