Sparse Bayesian Learning with Atom Refinement for mmWave MIMO Channel Estimation

Paper i proceeding, 2023

In this paper, we introduce a novel estimation method for the downlink millimeter-wave (mmWave) multiple-input multiple-output (MIMO) channel. The proposed method is able to determine the angles, time delays, and gains of the multi-path components by using the spatially sparse nature of mmWave channels. We first use on-grid sparse Bayesian learning (SBL) to coarsely estimate the channel parameters in the beamspace domain. We then develop a refinement method based on Newton-Raphson and Least Square-based atomic tuning to generate a mismatch-free basis. Finally, we finely reconstruct the channel by SBL using the basis found in the previous step. Simulation results show that the proposed channel estimation method outperforms the traditional ones in terms of mean square error and algorithmic complexity.