An online deep learning based channel estimation method for mmWave massive MIMO systems

Accurate channel estimation with low pilot overhead is one of critical tasks in hybrid millimeter wave (mmWave) massive multiple-input multiple-output (MIMO) system. Recently, deep learning (DL) algorithms have been developed to overcome this difficulty. The well trained neural network from a offline dataset infers the channel matrix from limited pilot symbols. However, this method suffers a potential performance penalty when the actual channel deviates from the pre-trained channel model. Thus, an online DL-based channel estimation framework for mmWave massive MIMO systems by leveraging the channel sparsity in the angular domain is proposed in this paper. In order to enable the network to converge without the need for real channel information, a label-free loss function and its convergence proof are given. Simulation results demonstrate that the proposed method achieved a better performance in accuracy and tracking tests than other existing approaches.