By eliminating conventional electronic processing, the system achieved a latency below 3 nanoseconds, thereby delivering more than a hundredfold improvement over traditional electronic sensing frameworks while maintaining high measurement accuracy.
Challenges in Traditional Optical Fiber Sensing
Optical fiber sensors are widely used in industrial monitoring due to their compact size, high sensitivity, and immunity to electromagnetic interference. These characteristics make them valuable for structural health monitoring, pipeline inspection, and more.
They operate by converting physical changes, including temperature variations and mechanical strain, into measurable changes in optical properties such as wavelength and intensity.
Despite these advantages, conventional optical fiber sensing systems face significant limitations. In traditional architectures, sensing and data analysis are performed separately, requiring optical signals to be converted into electrical signals through photodetectors. This optoelectronic conversion process increases power consumption and introduces latency, limiting real-time performance in large, densely multiplexed sensor networks.
Innovative All-Optical Processing Architecture
To eliminate the need for electronic signal processing, researchers developed a passive all-optical computing module based on a scattering medium and an optical diffraction network (ODN).
The new system directs the sensor output into a short segment of multimode fiber (MMF), which acts as a high-dimensional scattering medium. The MMF supports multiple propagation modes simultaneously, meaning even small variations in the sensor signal produce large yet predictable changes in the resulting speckle patterns.
In the experimental setup, signals from a fiber Bragg grating (FBG) sensor were routed through an optical circulator into the MMF using offset fusion splicing to enhance mode excitation.
The generated speckle field was then collimated, passed through polarization control elements, and projected onto a programmable spatial light modulator (SLM) implementing the ODN. For practical deployment, the SLM can be replaced with passive diffractive optical elements that require no electrical power.
An 80 mm focal length lens focused the processed optical field onto a detection plane, where a high-speed InGaAs photodetector or detector array measured the output. During system training, a genetic algorithm optimized the ODN phase-modulation pattern, creating a direct mapping between the sensor state and the detected optical intensity.
This approach enables direct signal interpretation within the optical domain, eliminating the need for conventional electronic processors and substantially reducing processing latency.
Want to save for later? Click here.
Experimental Validation and Performance Metrics
The AOFS-IC architecture demonstrated that optical domain computing can achieve high sensing accuracy while dramatically reducing processing latency. When coupled with an FBG sensor, the system successfully detected wavelength shifts as small as 3.3 pm.
The trained diffractive network maintained a linear response across different measurement ranges, achieving a strain resolution of 2.7754 με over a 150 με range and measuring strains up to 2.5 mε with a root-mean-square error (RMSE) of 0.0688 mε.
The platform also performed exceptionally well in discrete state classification. Using a 5 cm multimode fiber torsion sensor, the optical network classified rotational states in 45 ° intervals from 0 ° to 360 ° with 100% accuracy, even under low-signal conditions.
Additionally, the architecture demonstrated parallel sensing capability by simultaneously measuring strain and torsion within a single fiber, achieving RMSE values of 2.2829 με for strain and 0.7400 ° for torsion, with minimal detectable crosstalk between the two measurements.
For dynamic sensing, a single photodetector configuration achieved a static strain resolution of 1.6160 nε over a 95 nε range. The system accurately reconstructed acoustic and vibration signals up to 150 kHz while maintaining an ultra-low noise floor of 69 fε/√Hz over a 0-5 MHz bandwidth, indicating its robustness in detecting weak signals.
However, researchers acknowledged that the highest-frequency measurements were affected by photodetector bandwidth limitations.
Real-Time Applications in Robotic Systems
The low-latency and low-power characteristics of the AOFS-IC architecture make it well-suited for real-time edge computing in robotic systems.
To demonstrate its practical capabilities, researchers implemented a three-degree-of-freedom (3 DOF) joint-monitoring system on a JAKA industrial robotic arm. A single multimode fiber was routed across three rotating joints, using bending-induced mode coupling as the sensing mechanism.
The optical computing system translated the resulting wavefront distortions into three independent tracking regions, allowing simultaneous monitoring of multiple joint positions. For single-joint measurements, the system maintained a highly linear response across a 90 ° motion range with an RMSE of 1.5625 °.
During simultaneous three-joint motion, the AOFS-IC platform successfully separated overlapping optical signals and produced independent angle measurements with RMSE values of 1.7071°, 1.7003°, and 1.8755° for the three joints.
These results demonstrate the architecture's ability to perform real-time posture tracking and suggest its potential for closed-loop feedback control without the robot’s primary electronic processor.
Future Prospects for Photonic Integration
In summary, the AOFS-IC framework presents a high-performance alternative to conventional electronic signal interrogation systems. By performing sensing and signal processing directly within the optical domain, the architecture overcomes longstanding trade-offs between processing speed, sensor density, and power consumption.
The study highlights a clear path toward large-scale deployment through hardware miniaturization. Future implementations could integrate scattering media and passive diffractive elements into compact photonic chips, enabling robust, mass-producible sensors.
Such developments could support its use in industrial metrology, robotics, structural health monitoring, and automated manufacturing. Overall, the ability to process sensor information at optical speeds may also enable real-time monitoring of ultrafast events, including ballistic impacts, shock waves, and other high-frequency industrial processes.
Journal Reference
Tao, Y., et al. (2026). Nanosecond-latency all-optical fiber sensing with in-sensor computing. Light Sci Appl 15, 251. DOI: 10.1038/s41377-026-02265-x, https://www.nature.com/articles/s41377-026-02265-x
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.