AI enlightens wireless communication: A transformer backbone for CSI feedback

• DOI: 10.23919/jcc.ea.2022-0186.202401
• Publication Date: 2024-04-09T18:35:21Z
• Authors: Xiao Han; Wang Zhiqin; Li Dexin; Tian Wenqiang; Liu Xiaofeng; Liu Wendong; Jin Shi; Shen Jia; Zhang Zhi; Yang Ning
• Keywords: Signal Processing (eess.SP); 0202 electrical engineering, electronic engineering, information engineering; FOS: Electrical engineering, electronic engineering, information engineering; 02 engineering and technology; Electrical Engineering and Systems Science - Signal Processing

This paper is based on the background of 2nd Wireless Communication Artificial Intelligence (AI) Competition (WAIC) which hosted by IMT-2020(5G) Promotion Group 5G+AI Work Group, where framework eigenvector-based channel state information (CSI) feedback problem firstly provided. Then a basic Transformer backbone for CSI referred to EVCsiNet-T proposed. Moreover, series potential enhancements deep learning (DL-based) including i) data augmentation, ii) loss function design, iii) training strategy, and iv) model ensemble are introduced. The experimental results involving comparison between traditional codebook methods over different channels further provided, show advanced performance promising prospect DL-based problem.