Related Experiment Video
Updated: Dec 14, 2025
09:43
Transmission of Multiple Signals through an Optical Fiber Using Wavefront Shaping
Published on: March 20, 2017
10.2K
Advancing theoretical understanding and practical performance of signal processing for nonlinear optical
Qirui Fan1, Gai Zhou2, Tao Gui2
1Photonics Research Center, Department of Electrical Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong, China. remi.qr.fan@gmail.com.
Nature Communications
|July 25, 2020
Summary
Optimizing digital back propagation using deep neural networks improves optical communication by mitigating nonlinear effects and noise. This machine learning approach offers new insights into noise statistics for better signal processing.
Area of Science:
- Optical Communications
- Digital Signal Processing
- Machine Learning
Background:
- Compensating nonlinear effects in long-haul optical systems is challenging due to complex interactions between Kerr nonlinearity, chromatic dispersion (CD), and amplified spontaneous emission (ASE) noise.
- Standard digital back propagation (DBP) optimized as a deep neural network (DNN) shows promise in simulations for fiber nonlinearity compensation.
Purpose of the Study:
- To extend DNN-optimized DBP to practical optical communication experiments.
- To analyze the mathematical structure of optimized DNN-DBP parameters and investigate noise statistics in fiber nonlinearity compensation.
Main Methods:
- Experimental implementation of DNN-optimized DBP in single-channel and polarization division multiplexed wavelength division multiplexed systems.
- Analysis of noise statistics and distortions arising from ASE noise and incomplete CD compensation within DBP stages.
Main Results:
- Achieved improved performance compared to state-of-the-art DSP algorithms in practical experiments.
- Identified a trade-off between suppressing distortions and inverting fiber propagation effects across DBP stages.
- Revealed that ASE noise and incomplete CD compensation introduce cumulative distortions.
Conclusions:
- DNN-optimized DBP offers a superior approach for nonlinearity compensation in optical systems.
- Machine learning provides analytical insights into noise and distortions, advancing theoretical understanding beyond 'black-box' methods.
- Optimal DSP requires balancing distortion suppression and signal inversion, with this balance varying across DBP stages.
