Algorithmic Compensation of the Polarization Cross-Sensitivity of TFBG Sensors for SRI Measurement

Tilted fiber Bragg gratings are sensitive to a range of physical quantities, which unfortunately gives rise to undesirable cross-sensitivities. In refractive index measurements, the polarization of the radiation introduced into the optical fiber is an additional factor affecting the measurement. Analysis of optical spectra containing several dozen resonances mainly involves extracting features from the spectrum and constructing a regression model relating these features to the refractive index. In this article, we propose the use of a shallow artificial neural network as the regression model. The shifts in several dozen cladding modes were used as input features for the model. Additional features included the cutoff wavelength and the spectrum length, both of which were determined from the spectra. We demonstrated that the sensitivity of refractive index determination to polarization is reduced compared with classical methods such as the cutoff wavelength method.