Better Lightweight Network for Free: Codeword Mimic Learning for Massive MIMO CSI Feedback

The channel state information (CSI) needs to be fed back from the user equipment (UE) base station (BS) in frequency division duplexing (FDD) multiple-input multiple-output (MIMO) system. Recently, neural networks are widely applied CSI compressed feedback since original overhead is too large for massive MIMO Notably, lightweight attract special attention due their practicality of deployment. However, accuracy likely harmed by network compression. In this letter, a cost free distillation technique named codeword mimic (CM) proposed train better with practical encoder. A mimic-explore training strategy scheduler designed enhance CM learning. Experiments show that learning outperforms previous state-of-the-art method, boosting performance without any extra inference cost.