Massive MIMO Detection Method Based on Quasi-Newton Methods and Deep Learning
Yongzhi Yu, Shiqi Zhang, Ying Jie, Ping Wang
Abstract
Due to the increase in the number of antennas in massive Multiple-Input Multiple-Output (MIMO) systems, traditional MIMO detection algorithms need to be improved. In this letter, we introduce trainable variables in Broyden Quasi-Newton method to obtain Broyden-Net that avoids the high-dimensional matrix inversion of the linear Minimum Mean Square Error (MMSE) detector and also breaks through the performance limitations of the linear detector. To further simplify the complexity of the algorithm, we use the search direction consistency of the special form Broyden-Fletcher-Goldfarb-Shanno (BFGS) in Broyden to further combine the BFGS Quasi-Newton method with Deep Learning (DL), to propose BFGS-Net that is more adapted to massive MIMO detection. Numerical results show that Broyden-Net and BFGS-Net effectively reduce the computational complexity of the MMSE detector and can achieve good detection performance for massive MIMO systems in spatially correlated scenarios.