Search

Photonic crystal fiber-based polarization filter has important application prospects in optical fiber communications, nonlinear optics, and sensors. In order to achieve polarization filters with optimal performance, some numerical methods have been used to search for optimal results under different geometric structure parameters of fibers. However, it is unrealistic to calculate solutions under every parameter. In addition, the search direction and scope of adjusting fiber geometric structure parameters still depend on the designer's intuition and experience. Therefore, we propose a novel design method for photonic crystal fiber polarization filters to overcome these questions. The optimization design problem of polarization filters is transformed into the single objective function optimization problem. The combination of artificial neural networks and intelligent optimization algorithms realizes the optimal value-solving problem in two-dimensional parameter space. The method efficiently and automatically searches for optimal structures in the parameter space by fine-tuning both variables simultaneously, improving the bandwidth of the polarization filter and achieving higher crosstalk at 1550 nm. The method eliminates computationally expensive wavelength parameters and greatly reduces computation time. This novel design method realizes the reverse design of photonic crystal fiber-based polarization filters. In addition, the new method can also be used to design polarization filters at other objective wavelengths, and it is potentially applicable to the design of other photonic crystal fiber devices. • Wavelength parameters are removed, drastically cutting down on computation time. • The polarization filters optimization design is transformed into function optimization. • Machine learning techniques provided optimal design parameters of polarization filters. [ABSTRACT FROM AUTHOR]