GRU neural-network-assisted high-refractive-index sensing based on a no-core fiber with a waist-enlarged fusion taper structure

Shiwei Liu; Mengyuan Wu; Shuaihua Gao; Zhuang Li; Haoran Wang; Hongyan Fu

doi:10.1117/1.APN.4.4.046004

Advanced Photonics Nexus, Volume. 4, Issue 4, 046004(2025)

GRU neural-network-assisted high-refractive-index sensing based on a no-core fiber with a waist-enlarged fusion taper structure

Shiwei Liu¹, Mengyuan Wu¹, Shuaihua Gao¹, Zhuang Li¹, Haoran Wang^2、*, and Hongyan Fu^1、*

Author Affiliations

¹Xiamen University, National Model Microelectronics College, School of Electronic Science and Engineering, The Fujian Key Laboratory of Ultrafast Laser Technology and Applications, Department of Electronic Engineering, Xiamen, China

²Jimei University, School of Ocean Information Engineering, Fujian Provincial Key Laboratory of Oceanic Information Perception and Intelligent Processing, Xiamen, China

show less

Figures & Tables(13)

Fig. 1. (a) WEFT structure based on the NCF and (b) its microscopic photo.

Download full size

View in Article

Fig. 2. Optical field distributions of the WEFT structure based on the NCF with surrounding RIs at 1.430, 1.444, and 1.450, respectively.

Download full size

View in Article

Fig. 3. Experimental setup.

Download full size

View in Article

Fig. 4. (a) Spectral response of wavelength under different RIs and (b) two-dimensional spectral response.

Download full size

View in Article

Fig. 5. Polynomial fitting of wavelength resonance dip with RI.

Download full size

View in Article

Fig. 6. Structure of a single GRU cell.

Download full size

View in Article

Fig. 7. Schematic of GRU model architecture in this paper.

Download full size

View in Article

Fig. 8. Training curves of the GRU model for the proposed high-RI sensor based on NCF with WEFT structure: (a) loss curve and (b) RMSE curve.

Download full size

View in Article

Fig. 9. Comparison between true RI and predicted RI for GRU model-assisted high-RI sensor based on NCF with WEFT structure for the (a) training set and (b) test set.

Download full size

View in Article

Fig. 10. Error histogram between true RI and predicted RI for the test set.

Download full size

View in Article

Table 1. Fusion splicing parameters of WEFT structure based on NCF.
View table
View in Article
Table 1. Fusion splicing parameters of WEFT structure based on NCF.
Project Parameter
Prefusion time 180 ms
Prefusion power Standard
Distance $10 μ m$
Overlap $150 μ m$
Discharge time 2000 ms
Discharge intensity Standard

Table 2. GRU network model parameters.
View table
View in Article
Table 2. GRU network model parameters.
Layer Layer name Output shape Parameter
1 Input (800, 1, 1) —
2 Flatten (800, 1) —
3 GRU (64, 1, 1) 166,080
4 GRU (64, 1, 1) 24,768
5 GRU (32, 1, 1) 9312
6 GRU (16, 1, 1) 2322
7 GRU (4, 1) 252
8 FC^a (None, 1) 5
9 Output (None, 1) —

Table 3. Comparison of RI demodulation schemes.

View table

View in Article

Table 3. Comparison of RI demodulation schemes.


Structure	Demodulation technique	Performance	$R^{2}$	Ref.
TFBG	Residual CNN	$2.82 \times 10^{- 7}$ (MSE)	0.9982	28
TFBG-assisted SPR	Principal component analysis (PCA)	$9 \times 10^{- 6}$ (accuracy)	—	29
TFBG with FP	Double-branch CNN	0.0003174 (MAE)	0.9998	18
SMF-NCF-SMF	Artificial neural network (ANN)	0.000808 (MAE)	0.9957	30
NCF with WEFT	Wavelength drift	—	0.981	This work
GRU	0.00011 (MAE)	0.9993
$2.24 \times 10^{- 8}$ (MSE)

Tools

Get Citation

Copy Citation Text

Shiwei Liu, Mengyuan Wu, Shuaihua Gao, Zhuang Li, Haoran Wang, Hongyan Fu, "GRU neural-network-assisted high-refractive-index sensing based on a no-core fiber with a waist-enlarged fusion taper structure," Adv. Photon. Nexus 4, 046004 (2025)

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites