Deep Learning Based Channel Prediction for Cell-Free Massive MIMO Systems

This paper investigates the channel prediction model based on deep learning (DL) for cell-free massive multi-input multi-output (CF-mMIMO) systems. The prediction problems which utilizes both spatial and temporal correlation information are formulated. Combining the graph convolutional network (GCN) and the encoder of Transformer with the relative positional encoding, a novel spatial-temporal joint channel prediction model is proposed. A CF-mMIMO system is simulated and datasets are generated according to prediction problems. The hyper-parameters of the proposed model are determined by adjacent matrix generation, autocorrelation analysis and cross-validation. According to the spatial-temporal joint prediction performance evaluation, it is shown that the proposed model has higher prediction performance compared with the traditional DL models in the appropriate range of computational complexity.