Error-Robust Deep Learning-Based CSI Feedback in Massive MIMO Systems: A Multi-Rate Vector Quantization Approach

Junyong Shin, Yo-Seb Jeon

Abstract: This paper proposes an error-robust deep learning (DL)-based channel state information (CSI) feedback method for massive MIMO communication systems. The proposed method utilizes an autoencoder (AE) model integrated with a vector quantization module to support finite-rate CSI feedback. The proposed method also employs a codeword alignment module that not only provides error-robust codeword alignments but also facilitates a nested codebook structure to support multiple feedback rates using a single codebook. Simulation results demonstrate that the proposed method using a single AE model outperforms the existing method using multiple AE models in terms of CSI reconstruction performance.