Knowledge-Aided Deep Learning for Beamspace Channel Estimation in Millimeter-Wave Massive MIMO Systems
Millimeter-wave massive multiple-input multiple-output (MIMO) can use a lens antenna array to considerably reduce the number of radio frequency (RF) chains, but channel estimation is challenging due to the number of RF chains is much smaller than the number of antennas. To address this challenge, we propose a beamspace channel estimation scheme based on deep learning (DL) in this paper. Specifically, the beamspace channel estimation problem can be formulated as a sparse signal recovery problem, which can be solved by the classical iterative algorithm named approximate message passing (AMP), and its corresponding version learned AMP (LAMP) realized by a deep neural network (DNN). Then, by exploiting the Gaussian mixture prior distribution of the beamspace channel elements, we derive a new shrinkage function to refine the classical AMP algorithm. Finally, by replacing the activation function in the conventional DNN with the derived Gaussian mixture shrinkage function, we propose a complex-valued Gaussian mixture LAMP (GM-LAMP) network specialized for estimating the beamspace channel. The simulation results show that, compared with the existing LAMP network and other conventional channel estimation schemes, the proposed GM-LAMP network considering the channel knowledge can improve the channel estimation accuracy with a low pilot overhead.
READ FULL TEXT