End-to-end deep learning framework for coarsely sampled TFBG-SPR spectrum demodulation

Tilted fiber Bragg grating (TFBG) sensors integrated with surface plasmon resonance (SPR) are highly sensitive and versatile tools for various sensing applications. However, the complex spectral patterns produced by TFBG-SPR sensors pose challenges for accurate and efficient data interpretation. This study proposes a novel end-to-end deep learning framework for demodulating coarsely sampled TFBG-SPR spectra, achieving a refractive index resolution on the magnitude of -5 RIU. The framework enables real-time demodulation and supports cross-fiber adaptability, addressing the practical demands of diverse sensing environments. A compact, low-cost data acquisition system has been developed to showcase the model's industrial application potential. Experimental results demonstrate robust and accurate demodulation even under noisy and low-resolution conditions, highlighting the potential of this framework for cost-effective, portable, and high-accuracy TFBG-SPR sensing solutions.