Deep Learning-Based Time-varying Channel Prediction for MIMO Systems.

Shiyu Zhang,Yuxiang Zhang,Zhen Zhang,Jianhua Zhang,Liang Xia,Tao Jiang
DOI: https://doi.org/10.1109/vtc2022-spring54318.2022.9860388
2022-01-01
Abstract:Acquisition and feedback of accurate downlink (DL) channel state information (CSI) bring up high overhead. So a convolutional neural network (CNN) based multi-input-multi-output (MIMO) channel prediction method is proposed in this paper. Based on the space correlation, we design the proposed prediction method that predicts the MIMO channel concurrently. The proposed CNN method extracts the feature of UL-CSI through convolutional layers and predicts DL-CSI by the feature through deconvolutional layers. Simulation results demonstrate that the proposed channel prediction method leads to higher accuracy than existing methods. The bit error rate (BER) in orthogonal frequency division multiplexing (OFDM) system of proposed method is 82.4% and 26.2% lower than linear minimum mean squared error (LMMSE) method in two different simulated datasets. And the proposed method is tested under different user moving speeds and different antenna scale.
What problem does this paper attempt to address?