AcsiNet: Attention-Based Deep Learning Network for CSI Prediction in FDD MIMO Systems
Ya Jiang, Wenbin Lin, Weikun Zhao, Chaofeng Wang
Abstract
In 5G frequency division duplex (FDD) systems, the user equipment needs to feedback the measured downlink channel state information (CSI) to the base station to improve the throughput. For massive multiple-input-multiple-output (MIMO) systems, each antenna in base station needs its CSI feedback, which results in significant transmission overhead and latency. We propose an attention-based deep learning network to directly predict the downlink CSI from the corresponding uplink one, eliminating the feedback overhead completely. Specifically, the uplink CSI is first compressed based on the 3D inverse discrete Fourier transform, then is fed into an attention-based deep learning network which can focus on key CSI characteristics. The simulation results show that the proposed method achieves high prediction accuracy and low complexity, indicating prospective applications in FDD massive MIMO systems.