DeepChaos+: Signal Detection Quality Enhancement of High-Speed DP-16QAM Optical Fiber Communication Based on Chaos Masking Technique with Deep Generative Models

Abstract

In long-haul WDM (wavelength division multiplexing) optical communication systems utilizing the DP-16QAM modulation scheme, traditional methods for removing chaos have exhibited poor performance, resulting in a high bit error rate of 10−2 between the original signal and the removed chaos signal. To address this issue, we propose DeepChaos+, a machine learning-based approach for chaos removal in WDM transmission systems. Our framework comprises two key points: (1) DeepChaos+ automatically generates a dataset that accurately reflects the features of the original signals in the communication system, which eliminates the need for time-consuming data simulation, streamlining the process significantly; (2) it allows for the training of a lightweight model that provides fast prediction times while maintaining high accuracy. This allows for both efficient and reliable signal reconstruction. Through extensive experiments, we demonstrate that DeepChaos+ achieves accurate reconstruction of the original signal with a significantly reduced bit error rate of approximately 10−5. Additionally, DeepChaos+ exhibits high efficiency in terms of processing time, facilitating fast and reliable signal reconstruction. Our results underscore the effectiveness of DeepChaos+ in removing chaos from WDM transmission systems. By enhancing the reliability and efficiency of chaotic secure channels in optical fiber communication systems, DeepChaos+ holds the potential to improve data transmission in high-speed networks.

Keywords

• wavelength division multiplexing
• optical fiber communication
• machine learning
• variational auto encoder
• knowledge distillation