EVCsiNet: Eigenvector-Based CSI Feedback Under 3GPP Link-Level Channels
W. K. Liu, Wenqiang Tian, Han Xiao, Shi Jin, Xiaofeng Liu, Jia Shen
Abstract
Recently, deep learning based methods have been widely used for wireless communications. In this letter, different from current researches considering full channel state information (F-CSI) feedback, the eigenvector based CSI feedback with deep learning approach is proposed, referred to as EVCsiNet, where the joint eigenvector concatenated from multiple subbands is compressed and recovered at the encoder and decoder, respectively. Simulation results verify the superiority of our schemes over CSI recovery accuracy and feedback overhead compared with conventional methods using codebooks on link-level channel models in different scenarios.
Topics & Concepts
Channel state informationComputer scienceEncoderOverhead (engineering)Eigenvalues and eigenvectorsChannel (broadcasting)Joint (building)Link (geometry)WirelessAlgorithmComputer networkTelecommunicationsEngineeringPhysicsQuantum mechanicsArchitectural engineeringOperating systemWireless Signal Modulation ClassificationFull-Duplex Wireless CommunicationsAdvanced MIMO Systems Optimization