Flexible Link Adaptation in Fully-Decoupled RAN: A Machine Learning Approach.

Fully decoupled radio access network (FD-RAN), as an emerging radio access architecture through physical uplink-downlink decoupling and control-data decoupling, has potential in flexible spectrum utilization and network cooperation. However, real-time uplink feedback is challenging in FD-RAN due to the complete physical decoupling of control and data base stations. In this paper, we propose a flexible link adaptation mechanism that leverages outdated channel state information (CSI) to determine the appropriate Modulation and Coding Scheme (MCS) for the user in the FD-RAN downlink. Specifically, we first utilizes kernel recursive least squares to predict the CSI at the future moment. We then select the optimal modulation and coding scheme based on the predicted CSI and the frame error rate estimated by a neural network. Simulation results show that the proposed link adaptation mechanism has a significant throughput performance gain in various scenarios.