Abstract
Research on deep learning (DL)-based channel estimation for massive multiple-input multiple-output (MIMO) communication systems has attracted considerable interest in recent years. In this paper, we propose a DL-assisted channel estimation algorithm that transforms the original channel estimation problem into an image denoising problem, contrasting it with traditional experience-based channel estimation methods. We establish a new polarized self-attention-aided channel estimation neural network (PACE-Net) to achieve efficient channel estimation. This approach addresses the limitations of the conventional methods, particularly their low accuracy and high computational complexity. In addition, we construct a channel dataset to facilitate the training and testing of PACE-Net. The simulation results show that the proposed DL-assisted channel estimation algorithm has better normalization mean square error (NMSE) performance compared with the traditional algorithms and other DL-assisted algorithms. Furthermore, the computational complexity of the proposed DL-assisted algorithm is significantly lower than that of the traditional minimum mean square error (MMSE) channel estimation algorithm.