Channel Estimation for mmWave Massive MIMO Systems Using Graph Neural Networks

For massive multiple-input multiple-output(MIMO) systems, large-scale array antennas are used at both the base station and the receiver, which poses a challenging problem in channel estimation. Moreover, existing deep learning (DL) based channel estimation approaches generalize poorly when there is significant model mismatch between the train stage and the testing stage. In this paper, we propose a graph neural network (GNN) based channel estimation method by incorporating the graph topology of the system into the neural network architecture design. The mapping from the initial estimated channel matrix to the real channel matrix is learned by the proposed GNN, which inherits permutation equivariance and good generalization properties. Simulation results show that the proposed GNN based method generalizes well to different numbers of antennas at the user and maintains good performance when the number of antennas at the user increases dramatically.