随着多媒体宽带服务、物联网、5G以及人工智能等技术的发展，当今通信系统的骨干网物理层需要为急剧增长的数据量提供更高的传输带宽和更快的传输速度。相干光正交频分复用（CO-OFDM）以其高频谱利用率、高接收机灵敏度以及对色散容忍性强等优点成为物理层的关键技术。然而，CO-OFDM系统对时偏、频偏及相位噪声十分敏感，因此载波恢复对改善系统性能具有重要意义。针对CO-OFDM的以上问题，本项目自主开发一种新型基于机器学习的贝叶斯滤波技术来实现载波恢复。该技术通过不断更新后验概率以预测和追踪下一状态，可实现对时偏、频偏以及相位噪声的实时、动态、联合估计，从而大幅度降低接收机复杂度，并提升对其他信道损耗的容忍性。在理论推导及仿真分析后，将搭建高速实验平台进行验证。本项目的研究可建立光纤通信与机器学习算法间的桥梁，推动未来高速光纤通信快速发展，兼具学术及应用价值。

With the exponential increase in data traffic, especially the demand for ultra-broad bandwidth driven by multimedia applications, Internet of Things (IoT), 5G and Artificial intelligence (AI) technologies, ultra-high-speed optical fiber communication system has become the indispensable backbone of modern information infrastructures. Coherent optical orthogonal frequency-division multiplexing (CO-OFDM) has recently received much attention as a candidate for high-speed fiber transmission due to its key advantages: resilience to dispersion, and high spectral efficiency. However, the CO-OFDM system is highly sensitive to the timing offset (TO), carrier frequency offset (CFO) and laser phase noise (LPN). The carrier recovery is of great importance to improve the system performance. In this project, we propose to use Bayesian filtering (one of the key machine learning algorithms) for carrier recovery to ensure the optimal performance. The Bayesian filter updates the posterior probability to predict and track the state in the next time, which can achieve the real-time dynamic estimation of the timing offset, carrier frequency offset and the laser phase noise. Moreover, this method reduces the computational complexity of the receiver, while ensuring the joint estimation performance, and has an excellent tolerance to other distortions and noise in a real channel. After theoretical derivation and simulation analysis, high-speed experiments will be conducted to verify our results. The research of the project will build a bridge of optical fiber communication and machine learning, and improve the performance for our next generation high-speed optical communication systems. It has both academic impact and practical applications.