AI-Powered Communication Over Fiber-Optic Quasi-Distributed Acoustic Sensing Networks

Conventional distributed fiber-optic sensing (DFOS) systems are primarily limited to vibration, temperature, and strain monitoring, leaving many critical parameters reliant on discrete point sensors. In harsh environments, such as downhole, underwater, and underground, these point sensors face significant challenges due to the limited bandwidth, range, and robust ness of traditional communication methods. This work presents an AI-assisted communication framework that employs fiber optic quasi-distributed acoustic sensing interrogation to enable real-time data transmission from spatially distributed sensor nodes. The system employs a compact transmitter design and a direct-detection phase-sensitive optical time domain reflectometer (φ-OTDR) architecture, supporting both tightly coupled (piezoelectric-transducer-wrapped fiber) and spatially separated (speaker-offset) configurations. Two demodulation strategies are developed: a math-based baseline method and a lightweight convolutional neural network (CNN)-based approach. Experimental results demonstrate error-free data transmission at 4 kbps per sensing node using frequency-division multiplexing (FDM), with the ability to integrate multiple nodes simultaneously demodulated along a 1-km fiber. The CNN-based models achieve 100% classification accuracy across all FDM channels and spatial configurations. In a multi-node setup using single-carrier transmission, the CNN framework achieves 100% accuracy for three distributed nodes, outperforming the math-based baseline method, which yielded bit error rates (BERs) of 8.33 × 10−5, 5.83×10−3, and 8.33×10−5. Even under challenging conditions with physical decoupling, the CNN model maintains 99.8% accuracy, reducing the BER from 5.80 × 10−2. These results highlight the system's robustness, scalability, and suitability for next-generation distributed sensor networks—without requiring complex signal processing or hardware modifications.