Research on downlink channel state information prediction technique for 5G system based on deep neural network

In this paper, a deep learning-based downlink channel state information (CSI) prediction scheme is proposed, which utilizes the mapping relationship between the uplink channel and the downlink channel, based on a data-driven approach, mathematically proves and Orthogonal Frequency Division Multiplexing (OFDM) modulation pattern, to achieve the purpose of predicting the downlink CSI based on the uplink CSI. In order to make the deep neural network more suitable for processing high-dimensional CSI data, this paper designs a 3D-CsiNet network model, which uses 3D convolution to replace the traditional 2D convolution and improves the feature extraction and residual network modules. The results show that the 3D-CsiNet model proposed in this paper not only has high prediction accuracy, strong generalization ability, and higher accuracy of prediction performance while the number of parameters of the model is also lower, which is a more obvious advantage. In addition, the feedback performance of the 3D-CsiNet model works best when the total number of bits is set to 4. The binarization operation outperforms 4-bit quantization when the feedback bits are extremely limited, but is far worse in other cases. The original phase feedback method cannot feedback more useful information at NMSE > 0 dB, which is comparable to noise. The CSI phase feedback performance results in a performance gain of around 10 dB for BPD values of 0.5-0.6. This shows that the 3D-CsiNet network model proposed in this paper has high performance and efficiency in predicting downlink channel state information for 5G systems.