Fiber neural networks for the intelligent optical fiber communication signal processing

With the rapid growth of global data traffic, enhancing the intelligence and efficiency of optical fiber communication systems has become even more crucial. Current intelligent signal processing requires converting optical signals to electrical ones, leading to high latency and power consumption. To that end, a research collaboration led by Tsinghua University proposed a novel solution: a fiber neural network that processes information entirely within the optical domain.

“The system architecture consists of an input branch, an output branch, and an optical computing loop,” explains corresponding author Hongwei Chen. “Key components include a laser source, single-mode fiber, erbium-doped fiber amplifier, modulators, and dispersion-compensating fiber.”

The optical computing loop physically implements the linear operations of a neural network by using time-stretched optical pulses to perform vector-matrix multiplication—the core calculation between neural network layers.

“We applied the system to modulation format recognition, a fundamental task in optical communications. The fiber neural network successfully identified three modulation formats: OOK, PAM, and PSK,” shares Chen.

Notably, under ideal noise-free conditions, the system achieved 100% classification accuracy. It also maintained strong robustness when experimental noise from amplifiers and detectors was introduced, demonstrating its potential for real-world deployment.

The team's findings, published in iOptics, show that this fiber neural network framework can execute AI-driven computations directly using light.

“Importantly, this method avoids frequent optical-electrical conversions, thereby reducing processing delay and energy use,” adds Chen.

Beyond modulation recognition, the approach may be applicable to other intelligent optical communication applications such as fault detection and channel modeling, paving the way for deeper integration of artificial intelligence and photonic technology.

Contact author:

Hongwei Chen, Department of Electronic Engineering, Tsinghua University, Beijing, China. Email: chenhw@tsinghua.edu.cn

Funder: 

The Youth Fund of the National Natural Science Foundation (NSFC) of China under Grant 62301275, Key Laboratory of Radar Imaging and Microwave Photonics (Nanjing University of Aeronautics and Astronautics), Ministry of Education under Grant NJ20230003，Natural Science Research Startup Foundation of Recruiting Talents of Nanjing University of Posts and Telecommunications under Grant NY223032.

Conflict of interest: 

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

See the article:

Yubin Zang, Zuxing Zhang, Simin Li, Fangzheng Zhang and Hongwei Chen. Fiber neural networks for the intelligent optical fiber communication signal processing. iOptics (2025): 100009.Doi: https://doi.org/10.1016/j.iopt.2025.100009