Abstract
Optical fiber interferometric sensors are of great importance in chemistry, biology, and medicine disciplines owing to high-sensitivity and high-quality factor. However, due to the limitation of free spectral range, the inherent trade-off between wide measurement range and high sensitivity poses a persistent challenge in interference sensor development, which has fundamentally hindered their widespread adoption in precision measurement applications. In this work, a long short-term memory neural network is utilized in a Mach-Zehnder interference-based refractive index sensor to break the free spectral range limitation. Unique gating mechanism in long short-term memory neural network enables it to efficiently process long-term dependent sequence information, such as interference spectrum, avoiding the need for complex spectral signal analysis. A one-to-one mapping relationship is established between the interference spectrum and refractive index with root mean square error of 3.029 × 10(-4) and a coefficient of determination of 0.99971. The measurement range is extended from a single free spectral range of 1.3333-1.3561 to approximately three free spectral ranges of 1.3333-1.3921 without sacrificing sensitivity. Moreover, a wider measurement range can be achieved with sufficient training data. This work successfully resolves the inherent contradiction between high sensitivity and wide dynamic measurement range in optical interference-based sensors, opening up a path for the next generation of intelligent sensing systems.