HCC-Net: Holistic Cross-Joint Convolutional Network for CSI Feedback in Massive MIMO Systems

Abstract

With the rapid development of massive multiple-input multiple-output (mMIMO) techniques, the network capacity, number of served users and communication efficiency have been improved dramatically compared to that with limited number of antennas. These advantages are established based on accurate channel state information (CSI) at the base station (BS), which comes with a very high cost due to continuous CSI feedback from all the user equipments (UEs). In this letter, we propose a neural network-based CSI compression scheme with simple encoder-decoder framework for mMIMO systems. To achieve high accuracy, our proposed framework constructs an overall perceptual encoder-decoder structure with holistic cross-joint convolution (HCC) modules of different scales. In addition, a perceptual loss is introduced into the proposed design to further improve the accuracy in matrix recovery and limits the computational cost. Substantial experimental results demonstrate that the proposed HCC network (HCC-Net) is superior to several advanced algorithms in terms of estimation accuracy and computational complexity, such as the CSiNet+ and TransNet.