Automated Environment-Aware Channel Feedback for 6G Massive MIMO Systems

In the 5G Advanced and 6G, deep learning (DL)-aided channel state information (CSI) feedback has received much attention due to its ability to acquire an accurate downlink CSI at the base station (BS). Many of DL-based CSI feedback techniques are based on the autoencoder (AE), where the encoder is used to transform the CSI into a compressed codeword vector and the decoder reconstructs the original CSI from the codeword. A well-known drawback of these techniques is the absence of customized decoders capable of reconstructing CSI from diverse wireless environments. In this paper, we propose a novel environment-aware CSI feedback technique, referred to as automated environment-recognized CSI feedback (AER-CF) that addresses this limitation by recovering CSI with various channel characteristics. A key ingredient of AER-CF is to use automated environment recognition module (AERM), to intelligently group CSI with similar channel characteristics. In our work, we use AERM to identify the group to which the input CSI belongs and then train multiple decoders, each tailored for recovering CSI from a specific group. Numerical results demonstrate that AER-CF outperforms the conventional CSI feedback techniques in terms of CSI recovery performance.