image: Figure 1. LSTM-enabled interferometric sensing system with high sensitivity and wide measurement range. a. Sensing system mainly consists of a broadband light source, single-mode fiber (prepared based on the flame-based conical mechanism), and a spectral analyzer. Meanwhile, a schematic diagram of the internal sensing optical path is also shown in the figure. b. Dynamic spectral response and sensing mechanism. Conventional methods are limited to tracking spectral shifts within a single FSR. Changes in the refractive index (RI) across multiple FSRs introduce ambiguity in spectral response analysis. Specifically, within one FSR range, as RI changes, a new Dip D appears between Dip A and Dip B, and the spectrum overlaps with Dip C at the same wavelength, but they correspond to different measured values. This phenomenon persists across the entire measurement range, underscoring the need for intelligent spectral analysis. c. Intelligent data processing and feature learning process. To accurately decode complex spectral variations and map them directly to the value of RI, the system employs a LSTM network for end-to-end data processing. The network's memory cells can selectively retain spectral features relevant to the RI via the coordinated action of the input, forget, and output gates, thereby governing the update and output of the hidden state. d. Model performance and error analysis. The fully trained LSTM model exhibits extremely high prediction accuracy. The refractive index values predicted by the model are highly consistent with the actual measured values. This proves that the intelligent sensing system not only achieves high-precision refractive index measurement but also has excellent stability and reliability, providing a powerful solution for optical sensing in complex environments.
Credit: Junling Hu et al.
For a long time, optical interferometric sensors have been widely used in rapid and real-time detection in fields such as physics, chemistry, biology, and medicine owing to their unparalleled high sensitivity and high-quality factor. However, their limited free spectral range makes it difficult to achieve unambiguous monitoring over a broad measurement range while maintaining high sensitivity, which has significantly constrained their wider application in precision measurement scenarios. In recent years, researchers at home and abroad have attempted to integrate fiber Bragg gratings for spectral marking and track the changes of FSR with the measured quantity, etc., to alleviate these limitations. However, this often leads to complex manual calculations and other limitations that compromise sensitivity.
In a new paper published in Light: Science & Applications, a team of scientists, led by Professor Hailiang Chen from Shuguang Li 's group at Yanshan University, in collaboration with the team of Associate Professor Sigang Yang from Tsinghua University, innovatively proposed a full-spectrum recognition technology for optical interference spectra based on long short-term memory (LSTM). This method not only resolves the long-standing trade-off between high sensitivity and wide measurement range but successfully achieves both simultaneously.
To address this challenge, the research team integrated LSTM neural networks with fiber-optic interferometric sensing to overcome the limitation imposed by the FSR on the measurement range. Unlike traditional methods confined to a single spectral cycle, this technology leverages the gating mechanisms and sequence learning capabilities of LSTM to model long-term dependencies in complex interference spectra. It enables accurate identification even with spectral overlap, tripling the detection range while maintaining high sensitivity. This achieves a synergistic enhancement in both wide dynamic range and high precision.
Another key innovation is an efficient down-sampling strategy. By optimizing spectral sampling points, it significantly reduces data acquisition and processing loads without compromising measurement accuracy. This greatly improves system response speed, laying the foundation for practical use in dynamic, rapid-detection scenarios.
This technology uses algorithms to push hardware beyond its conventional physical limits, advancing optical sensing toward more intelligent and practical applications. By establishing an end-to-end system that maps interference spectra directly to target quantities, it circumvents the limitations of manual feature extraction. This provides a novel technical pathway for intelligent monitoring in biochemical and medical sensing within complex environments.
Journal
Light Science & Applications
Article Title
LSTM-assisted optical fiber interferometric sensing: breaking the limitation of free spectral range