MRFNet: A Deep Learning-Based CSI Feedback Approach of Massive MIMO Systems
Zhengyang Hu, Jianhua Guo, Guanzhang Liu, Hanying Zheng, Jiang Xue
Abstract
In frequency division duplex (FDD) networks, channel state information (CSI) is critical for massive multiple-input multiple-output (MIMO) systems, because the base station (BS) relies on accurate CSI to achieve the promising benefits of massive MIMO systems. Therefore, the user equipments (UEs) need to feedback CSI to the BS accurately. Since the feedback overhead is large in massive MIMO system due to the large number of antennas, compressing and recovering CSI more accurately are essential and urgent problems. In this letter, we propose a network named MRFNet. The MRFNet can recover fruitful features with different receptive fields and the large number of convolution channels for better CSI recovery. Moreover, we visualize the output of the last block and the final output of the MRFNet with a different number of convolution channels and multiple receptive fields to explain why MRFNet works. Simulation results show that the proposed MRFNet obtains the significant performance gain of CSI feedback for both indoor and outdoor scenarios with different compression rates.