Block Sparse Channel Estimation based on Residual Difference and Deep Learning for Wideband MmWave Massive MIMO

Time-domain channel estimation for wideband millimeter wave (mmWave) MIMO OFDM systems is considered. To mitigate the overfitting of the existing time-domain channel estimation exploiting block sparsity (TDCEBS) scheme, we propose a block sparse channel estimation exploiting residual difference (BSCERD) scheme, where we first compute the difference of the residual power for every two adjacent iterations, and then determine a threshold to indicate the convergence of the iterations. Moreover, to improve the global optimality and reduce the time overhead of compressive sensing, a block sparse channel estimation based on deep learning (BSCEDL) scheme is proposed to determine the indices of the nonzero blocks simultaneously. We exploit the QuaDRiGa to assess the efficacy of the schemes proposed. Simulation results show that both BSCERD and BSCEDL outperform TDCEBS, while BSCEDL is better than BSCERD in performance and can achieve much lower time overhead.