End-to-end Deep Learning of Optical Fiber Communications

Abstract: In this paper, we implement an optical fiber communication system as an end-to-end deep neural network, including the complete chain of transmitter, channel model, and receiver. This approach enables the optimization of the transceiver in a single end-to-end process. We illustrate the benefits of this method by applying it to intensity modulation/direct detection (IM/DD) systems and show that we can achieve bit error rates below the 6.7\% hard-decision forward error correction (HD-FEC) threshold. We model all componentry of the transmitter and receiver, as well as the fiber channel, and apply deep learning to find transmitter and receiver configurations minimizing the symbol error rate. We propose and verify in simulations a training method that yields robust and flexible transceivers that allow---without reconfiguration---reliable transmission over a large range of link dispersions. The results from end-to-end deep learning are successfully verified for the first time in an experiment. In particular, we achieve information rates of 42\,Gb/s below the HD-FEC threshold at distances beyond 40\,km. We find that our results outperform conventional IM/DD solutions based on 2 and 4 level pulse amplitude modulation (PAM2/PAM4) with feedforward equalization (FFE) at the receiver. Our study is the first step towards end-to-end deep learning-based optimization of optical fiber communication systems.

• The paper introduces a deep neural network that integrates the entire transceiver chain for IM/DD optical systems.
• The approach achieves bit error rates below the 6.7% HD-FEC threshold and supports 42 Gb/s data transmission over distances above 40 km.
• The research employs an adaptive training methodology that outperforms traditional PAM2/PAM4 modulation methods for diverse dispersion scenarios.

End-to-End Deep Learning of Optical Fiber Communications

This paper addresses the novel implementation of an optical fiber communication system as a comprehensive end-to-end deep neural network, embracing the complete transmitter, channel model, and receiver chain. The central focus is on optimizing the transceiver as an integrated, single process, particularly for intensity modulation/direct detection (IM/DD) systems.

Contemporary optical fiber systems face significant obstacles due to chromatic dispersion and the nonlinear Kerr effect, which constrain information rates. The application of artificial neural networks (ANNs), especially multi-layer frameworks that facilitate deep learning, has demonstrated potential in compensating for these impairments through channel equalization. The authors leverage this capacity to construct a full system learning paradigm that transcends conventional modular processing by designing communication networks via end-to-end training, promising optimized performance across the system.

The research demonstrates the system's capability to reach bit error rates below the 6.7% HD-FEC threshold, experimentally verifying information rates of 42 Gb/s over distances exceeding 40 km. This utilizes a sophisticated ANN model encompassing the transmitter and receiver, integrated within the nonlinear channel model of the optical fiber. The proposed communication network is modeled as a deep, fully-connected feedforward ANN where optimizations target both symbol error rate reductions and achieving robust transceivers that maintain reliable transmission across various fiber dispersions without requiring reconfiguration.

The authors' approach significantly outperforms traditional IM/DD solutions employing PAM2/PAM4 modulation with feedforward equalization, highlighting the advantages of a non-modular, integrated design strategy. Furthermore, the paper emphasizes a robust training methodology that avails flexible transceivers adept at working over a wide array of dispersion scenarios, spearheaded by simulations validated through experimental application.

This end-to-end approach marks the inception of transforming optical fiber systems through deep learning mechanisms, representing a significant shift in design principles towards employing holistic optimization strategies. The implications are substantial—an end-to-end system as opposed to a system of discreet optimized components; this restructuring facilitates optimized data throughput, reduced error rates, and adaptation across varied operational conditions.

Looking ahead, this research may inspire further applications in diverse optical communication contexts, potentially affecting future transceiver designs not restricted to IM/DD systems. Future research may expand on this foundation, addressing complexities inherent in more intricate communications channels and exploring additional layers of ANN integration. Potentially, this could encompass unsupervised training dynamics extending beyond supervised approaches, achieving broader adaptability across differing operational landscapes.

The paper lays groundwork towards a potential overhaul in optical communication design philosophy, from optimization at individual stages to embracing deep-learning based systemic approaches—highlighting an era whereby traditional barriers may be mitigated through innovative end-to-end artificial intelligence implementations.