The paper introduces a finite-rate deep-learning-based channel state information (CSI) feedback method for massive MIMO systems. It leverages vector quantization to provide a finite-bit representation of the latent vector, reducing computational complexity. The proposed method improves CSI reconstruction performance while minimizing feedback overhead. By separating the shape and gain of the latent vector, the approach significantly reduces computational complexity compared to conventional methods. Multi-rate codebook design enhances performance under varying feedback overheads.
In eine andere Sprache
aus dem Quellinhalt
arxiv.org
Wichtige Erkenntnisse aus
by Junyong Shin... um arxiv.org 03-13-2024
https://arxiv.org/pdf/2403.07355.pdfTiefere Fragen