Data-Driven Inverse Design Method for Photonic Crystal Fibers and Its Application to Near-Infrared Flat Supercontinuum Control

DOI: https://doi.org/10.62517/jike.202604311

Author(s)

Lanxin Wei

Affiliation(s)

Mudanjiang Normal University, Mudanjiang, Heilongjiang, China

Abstract

To overcome the issues of traditional photonic crystal fiber (PCF) design methods, such as the need for trial-and-error, low efficiency, and inability to adequately meet diverse spectral requirements, this paper proposes a big data-driven inverse design method. This method inversely derives the optimal PCF structural parameters based on a target spectrum, thereby enabling precise control to achieve a flat supercontinuum (SC) in the near-infrared (NIR) range. A large number of different structures are simulated using the finite element method (FEM) to generate a "structure–property" sample set. This dataset is then used to train a deep neural network (DNN) as a forward surrogate model, capable of computing the dispersion curve, nonlinear coefficient, and other parameters within milliseconds. Subsequently, this surrogate model is coupled with particle swarm optimization (PSO) to find the ideal PCF structure, with the objectives of minimizing spectral flatness and maximizing spectral bandwidth. Finally, the nonlinear Schrödinger equation (NLSE) is employed to inversely design the supercontinuum generation process within the fiber, and to investigate the control mechanisms of soliton self-frequency shift (SSFS), dispersive waves (DWs), as well as the robustness of structural parameters. The expected outcome is the establishment of a workflow encompassing "simulation data generation → neural network training → inverse design optimization → performance testing", resulting in a supercontinuum with good flatness (e.g., within ±3 dB) over a broad NIR range (e.g., 1400–1700 nm). This work opens a new route for designing high-performance PCFs and lays a solid foundation for further development of applications based on supercontinuum sources.

Keywords

Photonic Crystal Fiber (PCF); Inverse Design; Data-Driven; Supercontinuum (SC); Neural Network

References

[1] Ranka J K,Windeler R S & Stentz A J.(2000).Visible continuum generation in air-silica microstructure optical fibers with anomalous dispersion at 800 nm..Optics letters,25(1),25-7. [2]Dudley JM,Genty G & Coen S.(2006).Supercontinuum generation in photonic crystal fiber.Reviews of Modern Physics, 78(4),1135-1184. https://doi.org/10.1103/RevModPhys.78.1135. [3] Xu J W, Xu J J, Guo T X & Du B. Progress in application research of supercontinuum laser. Chinese Journal of Quantum Electronics, 1-23. (in Chinese; original journal, article in press) [4] Gong H T, Zhang B & Hou J. (2026). Research progress of visible to mid-infrared fiber supercontinuum. Journal of National University of Defense Technology, 48(02), 200-213. [5] Jian D, Liu M, He D D, Li D & Liao Z Y. (2013). Study on highly nonlinear photonic crystal fiber with flattened dispersion. Laser Technology, 37(02), 187-190. [6] Yuan J H, Hou L T, Zhou G Y, Gao F & Wang K. (2008). Fabrication and experiment of flattened dispersion highly nonlinear photonic crystal fiber. Optoelectronics • Laser, (08), 1007-1010. https://doi.org/10.16136/j.joel.2008.08.010. [7] Jiang G Y, Xiong W L, Fu Y J & Zhu Q S. (2012). Characteristics of supercontinuum generation in highly nonlinear photonic crystal fibers. Chinese Journal of Quantum Electronics, 29(01), 89-95. [8] Gao P F. (2017). Numerical study on mid-infrared supercontinuum generation based on highly nonlinear fibers (Master’s thesis). Shaanxi Normal University. https://kns.cnki.net/kcms2/article/abstract?v=... (URL truncated as in original) [9] Song K L. (2024). Optimal design of infrared emission spectra and multi-band control of metamaterial structures (PhD dissertation). Harbin Institute of Technology. https://doi.org/10.27061/d.cnki.ghgdu.2024.002708. [10] Li Y. (2020). Numerical study on supercontinuum generation based on highly nonlinear fibers (Master’s thesis). Xiangtan University. https://doi.org/10.27426/d.cnki.gxtdu.2020.001458.