Exploiting the Clustered Sparsity for Channel Estimation in Hybrid Analog-Digital Massive MIMO Systems

Although the hybrid analog–digital processing reduces the hardware complexity to realize massive multiple-input multiple-output system, it increases the difficulty of channel estimation significantly since the base station cannot obtain enough training samples with the limited number of radio frequency chains. This paper aims to address this problem using the emerging machine-learning techniques. By exploiting the sparsity property of the angular domain channel, a novel clustered sparse Bayesian learning approach is proposed to estimate the channel reliably. The scheme does not require the knowledge of channel statistics and angular information of users. A data-aided channel estimation scheme based on variational optimization is proposed to further improve the estimation performance by using the training data and unknown information data jointly. The numerical simulations show that the proposed schemes outperform the reference schemes significantly in terms of the normalized mean square error and the achievable sum rate.