Calculation Of Pulse Performance In Mode-locked Fiber Lasers And Multimode Fibers Based On Machine Learning

With the development of fiber optics and ultrafast pulse technology,mode-locked fiber lasers are widely used in scientific research and production due to their good performance in peak power,beam quality,heat dissipation,and stability,and related research has significant implications for fields such as communication,medicine,sensing,industry,and materials.As the application requirements increase,the pulse quality of mode-locked lasers also needs to be correspondingly improved,so how to produce high-quality pulses from mode-locked fiber lasers has gradually become an important research topic.By studying the relationship between cavity parameters and output pulses,people have gained new insights into the physical properties and evolution behavior of mode-locked pulses,which makes it possible to achieve specific pulse output by adjusting cavity components.However,due to the speed limitations of algorithms,related numerical studies are usually confined to one-dimensional or twodimensional cavity parameter spaces.Meanwhile,in experimental work,researchers often build lasers based on experience and established conclusions,which inevitably leads to a lot of trial and error and can waste time and economic costs.On the other hand,the application of high-quality pulses is closely related to their transmission process.Large-mode-field multimode fibers have a higher damage threshold and more capacity for propagation modes than single-mode and few-mode fibers,and have gradually become an important carrier for high-power pulse transmission.However,due to the complex nonlinear interactions between modes,it is relatively difficult to control the optical field in multimode fibers.In this context,this paper proposes using classification and regression models in the field of machine learning to construct a mapping between cavity parameters and output pulses,thereby achieving rapid traversal of pulse information in high-dimensional cavity parameter space.Combining with search algorithms,the inverse design of mode-locked fiber lasers is realized,and the search algorithm is also used to calculate the incident optical field solution when a special self-cleaning phenomenon occurs in multimode fibers,providing a new research method for optical field control in multimode fibers.The main research results are as follows:1.Based on the artificial neural network(ANN)model,the mapping relationship between the output pulse and cavity parameters of the Fabry-Perot pulse laser is established.This ANN has an accuracy of over 98%on the sample set and in actual calculations,and the single operation time is only about 10ms.Based on this,the convergence domain pattern in the three-dimensional cavity parameter space is calculated.An ANN that can directly output the lock mode pulse waveform has been trained and its accuracy has been verified in practical use.2.The pulse convergence types in complex mode-locked lasers are defined as strong convergence and generalized convergence.Support vector machines(SVMs)that can be used to distinguish these two pulse convergence types in nonlinear loop mirror(NOLM)mode-locked fiber lasers have been trained.The SVMs used to determine strong and generalized convergence are called SVMs and SVMg,respectively.The accuracy of SVMs and SVMg in instance calculations is 94.58%and 88.36%,respectively,and the single calculation time is about 8ms.Using them,the influence of each cavity parameter on the mode-locked pulse of the NOLM laser is analyzed.Two neural networks have been trained to output the time-domain and spectral patterns of the NOLM mode-locked fiber laser output pulse,and their regression coefficients R on the sample set exceed 0.99.They can accurately and quickly predict the time-domain and spectral characteristics of the pulse in practical use.3.Based on the trained machine learning models and genetic algorithms,an efficient laser reverse design algorithm has been designed,which can calculate the parameter values of the laser in reverse given the known pulse output information.Using this algorithm,the construction of fully polarization-maintaining fiber pulse flat-top lasers,100 MHz repetition rate fiber lasers,and parabolic spectrum NOLM modelocked fiber lasers has been achieved.4.The incident light field when the LP11 mode self-cleaning phenomenon occurs in graded-index multimode fiber(GIMF)is calculated using a search algorithm,and the numerical results show that when self-cleaning occurs,over 62%of the pulse energy converges to the LP11 mode.The numerical stability of this solution is analyzed,and an algorithm for solving the incident light field when considering more transmission modes is given.The main innovations are as follows:1.Proposing using ANN and S VM to analyze the pulse convergence problem in modelocked fiber lasers and calculates the convergence domain pattern in highdimensional cavity parameter space.Using ANN achieves high accuracy prediction of lock mode pulse related information.2.An efficient laser reverse design algorithm has been designed,and specific pulse width,repetition rate,and spectral shape lasers have been made using it.This algorithm provides convenience for the design of fiber lasers and is expected to play an important role in future laser design work.3.By combining the Rosenbrock search algorithm and genetic algorithm,the incident light field that produces stable higher-order mode self-cleaning phenomenon in GIMF is calculated,providing a new numerical research method for the study of light field control in GIMF.