AI-based CSI Prediction for 5G-Advance and 6G Networks

This article develops a novel Artificial Intelligence (AI)-based Channel State Information (CSI) prediction framework via Long Short-Term Memory (LSTM) networks for 5G-A and 6G. To rigorously assess the framework's performance, we conduct comprehensive performance simulations that incorporate realistic imperfections, such as channel estimation inaccuracies, to evaluate CSI prediction accuracy and the generalization capabilities of our model. Additionally, we undertake field trials to validate the framework's efficacy in operational networks. The simulation outcomes and field evaluations conclusively demonstrate that the AI-driven CSI prediction surpasses Wiener-based methods in terms of prediction accuracy. Furthermore, the simulations reveal strategies to mitigate AI model generalization challenges, notably through the utilization of mixed datasets. Ultimately, grounded in these insights, we formulate deployment recommendations for AI-based CSI prediction in 5G-Advanced and 6G networks.