A Deep Learning-Based CSI Prediction Method for LiFi Systems

Light-fidelity (LiFi) is a bidirectional, high-speed, and fully networked optical wireless communication technology, which requires the downlink channel state information (CSI) to enhance resource allocation and scheduling at the access point (AP). The downlink CSI can be acquired via the feedback signal at the cost of the uplink throughput degradation. In this work, for the first time we propose a deep learning-based CSI prediction method for LiFi systems, in which a quantile regression convolutional gate recurrent unit network (QR-ConvGRU) is used to forecast the downlink CSI based on the uplink CSI at the AP without any feedback overhead. Experimental results show that the QR-ConvGRU achieves improved prediction accuracy performance than fully connected neural network and ConvGRU as well as demonstrating robustness across different communication link spans in an indoor environment. The proposed QR-ConvGRU provides an innovative solution in the field of LiFi CSI prediction and contributes a new prediction framework for wireless communication.