An Edge-Deployable Intelligent Measurement Framework With Uncertainty Quantification and Interpretability for Submillisecond Demodulation in Plasmonic TFBG Sensors

Combining compact design with nanoscale surface refractive index (RI) detection capabilities, surface plasmon resonance (SPR) based on tilted fiber Bragg gratings (TFBG) offers a powerful platform for sensing in diverse fields such as battery health monitoring and point-of-care diagnostics. However, current measurement systems rely on high–spectral-resolution spectrometers and complex calibration, which limit their deployment in practical environments. This paper presents a physics-grounded intelligent measurement framework that directly demodulates sparsely sampled plasmonic spectra acquired by a low-cost CMOS detector. By integrating an end-to-end deep learning architecture with uncertainty quantification and interpretability, the framework enables reliable and traceable quantification of RI variations without high-resolution instrumentation. The proposed system achieves a RI resolution of 1.25e-5 RIU and a standard error of 4.16e-6 RIU, verified on both trained and unseen sensor geometries. Real-time implementation on an ARM Cortex-A78AE processor demonstrates sub-millisecond demodulation (0.8 ms latency, 1 kHz rate) with high stability. These results highlight a pathway toward miniaturized, uncertainty-aware TFBG-SPR instrumentation for the applications in emerging energy and advanced clinical diagnostics.