Artificial intelligence-enabled link adaptation

Abstract

Signaling resource overhead associated with current communication link adaptation mechanisms can be quite large and such mechanisms typically rely upon a channel state information (CSI) feedback process that can result in poor scheduling performance. Embodiments are disclosed in which a first device channel state information characterizing a wireless communication channel between the first device and a second device, and trains a machine learning (ML) module of the first device using the CSI as an ML module input and one or more modulation and coding scheme (MCS) parameters as an ML module output to satisfy a training target. By applying the concepts disclosed herein, overhead associated with feedback for MCS selection may be reduced compared to conventional link adaptation procedures, because, once ML modules at a pair of devices have been trained, the MCS selection by the ML modules can be done without requiring the ongoing feedback of CSI.