A Deep Neural Network-based Method for MmWave Time-varying Channel Estimation

A timevarying channel model makes estimating the channel coefficients challenging for the millimeter wave (mmWave) multi user multi-input multi-output (MIMO) communication, attributable to the many coefficients that have to be estimated with a limited number of measurements as well as the severe propagation loss experienced by the mm Wave band. Thus, it is proposed to divide the channel estimation in time-varying mm Wave systems in two stages, using a frame structure and assuming that angles of arrival/departure (AoA s/AoDs) vary much more slowly than path gains. Mm Wave channels have a sparse nature that is leveraged in the first stage to formulate the estimate of AoAs/AoDs as a block-sparse signal recovery problem. By the obtained estimate of the AoAs/AoDs, in the second stage the beamforming that maximize the desired pilot power is utilized in order to measure the path gains accurately. In this article, we propose the Deep Neural Network based Angle Estimation (DNNAE) algorithm by defining a deep neural network structure with appropriate input and output. Accordingly, we provide a method based on machine learning to increase the accuracy of channel AoDs/AoAs estimation. Therefore, without the need to update the angle grid area and with low complexity, we obtain a suitable estimation accuracy. Simulation results demonstrate that with the proposed DNNAE scheme, we outperform the previously proposed Adaptive Angle Estimation (AAE) algorithm despite the much lower computational complexity.