Demonstration of Intelligent Hybrid FSO/RF System Based on Enhanced GRU Prediction and Real-World Meteorological Dataset

Hybrid free space optical (FSO)/radio frequency (RF) system has emerged as a high-data-rate solution for the last mile access network. In this paper, we establish a hybrid FSO/RF testbed including an intensity modulation/direct detection (IM/DD) FSO subsystem with space diversity and a high-order quadrature amplitude modulation (QAM) RF subsystem with polarization diversity, respectively. Based on the testbed, we propose and experimentally validate an intelligent transmission system where the link selection and switching technology is implemented. In particular, we propose a gated recurrent unit (GRU) neural network enhanced by time attention mechanism for the hybrid system. The experimental results demonstrate that the proposed system can achieve the prediction of FSO channel fading with a high precision where absolute percentage error (APE) values lower than 3% account for up to 90% of the prediction results. The implemented intelligent hybrid system can significantly reduce the link switching frequency and link interruption duration. Then the BER is further verified on the testbed, and results show that the intelligent hybrid system can improve the BER from the order of 10−2to the order of 10−3during the time periods when link switching occurs.