Learning-assisted full-spectrum refractive-index interrogation for a plasmonic photonic crystal fiber sensor modeled by the finite element method
Abstract
Reliable refractive-index (RI) interrogation in photonic-crystal-fiber surface-plasmon-resonance (PCF-SPR) sensors is commonly performed by resonance-wavelength tracking, which uses only one spectral feature and may become unstable under broad peaks, spectral distortion, or measurement perturbations. In this work, a heptagonal PCF-SPR sensor modeled by the finite element method (FEM) is used as the physical sensing platform, and a learning-assisted full-spectrum interrogation framework is proposed to estimate analyte RI from the complete confinement-loss profile. The sensor combines an asymmetric heptagonal air-hole lattice, an open analyte channel, and an inner gold coating to enhance coupling between the fundamental core mode and the surface plasmon polariton mode. The optimized structure operates over an RI range of 1.31–1.42 and achieves a maximum wavelength sensitivity of 12 400 nm per RI unit (RIU). To reduce the cost of repeated FEM spectral scanning, a Fourier-enhanced residual surrogate is trained using spectrum-level train/validation/test partitioning. On the independent test set, the surrogate achieves a global root-mean-square error of 21.35 dB cm −1, a global coefficient of determination ( R2) of 0.87, and a mean peak-wavelength error of 6.7 nm. The trained surrogate is then embedded into a compensated full-spectrum inverse module, where candidate RI spectra are generated and matched with a target spectrum. Under clean-condition testing, the method achieves an RI mean absolute error of 7.91×10−4 RIU, outperforming conventional peak-tracking and spectral-matching baselines. These results demonstrate that full-spectrum interrogation can improve the numerical robustness of RI readout by using peak position, linewidth, slope, amplitude variation, and overall spectral morphology simultaneously. Experimental fabrication and calibration remain necessary for practical validation.