Abstract
A dual-input silicon-on-insulator (SOI) arrayed waveguide grating (AWG)
fiber Bragg grating (FBG) interrogator, enabled by an on-chip
Mach–Zehnder interferometer (MZI) switch, is proposed and
experimentally demonstrated. The device integrates a polarization beam
splitter (PBS), a thermo-optic MZI, and an asymmetric dual-input AWG.
The MZI enables on-chip selection of two AWG input ports, generating
complementary spectral responses without increasing the number of
output channels. Due to fabrication-induced non-idealities in SOI-AWG
responses, conventional fitting-based interrogation methods suffer
from increased errors. A feature-engineered feedforward neural network
(FNN) is developed by exploiting multichannel responses from the two
input states. Experimental results show that the proposed dual-input
FNN achieves an interrogation root-mean-square error (RMSE) of
2.33 pm on an independent external test set, reducing the error
from 8.06 pm obtained by polynomial fitting. This work provides
a compact and high-precision solution for on-chip FBG
interrogation.