Intelligent analog radio over fiber aided C-RAN for mitigating nonlinearity and improving robustness

As a low-cost solution for the 5G communication system, centralised radio access network (C-RAN) has been implemented in the ultra-dense environment, where radio over fiber (RoF) technology can enable reduced operational cost as well as coordinated multi-point (CoMP) despite its less robustness and reduced system performance. On the other hand, machine learning has been recognised as an efficient method for accelerating the fiber-optic communications with the aid of the advancements of the learning algorithms as well as the available high processing capabilities. In this paper, we propose a supervised learning-aided A-RoF system, where the logistic regression classification is invoked for removing the ARoF module’s need for re-customization and for boosting its performance. As a result, we can adaptively select the modulation format according to the optical power and the RF voltage, where we obtain an enhanced spectral efficiency and dynamic range (DR) by a factor of 4/3 and 19/13, respectively, while the learning network can be updated online