Two-Sided CSI Prediction Mechanism for Low-Overhead Control Feedback in 5G and Beyond

Abstract

In its Release 18, the 3rd Generation Partnership Project (3GPP) started exploring the benefits of leveraging artificial intelligence for enhanced performance and reduced overhead of the New Radio air interface. The Release 18 study item focused on Channel State Information (CSI) feedback, beam management, and positioning accuracy. Alongside CSI compression, time-domain CSI prediction at the user’s side was selected as a representative use case to combat the channel aging issue. In this work, we propose a two-sided, i.e., both at the user’s and the base station’s sides, machine learning-based CSI prediction solution to reduce uplink overhead without compromising the downlink performance. The proposed approach is evaluated using system-level simulations, which showed that our solution can reduce up to 30% of the CSI feedback overhead while enhancing the 5th percentile user-perceived throughput by up to 4% gain over the baseline zero-order hold scenario used in legacy cellular systems.