Dynamic channel model estimation based on gradient descent method and its optimization in massive MIMO

In dynamic environments, rapid changes in wireless channels make MIMO channel estimation and beam assignment based on traditional communication algorithms a great challenge. In this study, we try to solve the dynamic channel model by approximate inference and propose a method based on Stein’s variational gradient descent. By taking advantage of the low-rank nature of the massive MIMO channel matrix, the channel estimation problem is modeled as a variational inference optimization problem, and the SVGD algorithm is used to optimize the channel state information of the users. A dynamic channel simulation method is proposed to dynamically configure the FPGA according to the complexity of static, dynamic and born-again channels. The results show that the beamforming capability of this paper’s method can be stabilized at a gain of 35.5, and the simulation error of SVGD’s dynamic channel model is overall lower than 0.05 and has high robustness. The experimental results prove the feasibility and practicability of this paper’s method.