基于数字孪生信道的大规模MIMO信道低开销预测研究OA

Accurate Channel State Information(CSI)is critical for optimizing the performance of massive Multiple Input Multiple Output(MIMO)systems.However,the large number of antennas in these systems results in significant overhead and latency for CSI acquisition.This paper proposes a novel channel prediction method based on a Crossformer network,leveraging digital twin channels to achieve precise and efficient CSI prediction,thereby guiding the physical channels accurately.The method utilizes historical CSI data from multiple previous time slots and terminal position coordinates in the twin environment as inputs to the neural network.Initially,a dimensional segmentation module embeds the multivariate time series inputs into a two-dimensional vector array to preserve temporal,spatial,and cross-dimensional information.Subsequently,a second-order attention layer captures dependencies across multiple dimensions.Finally,a hierarchical encoder-decoder structure exploits the spatiotemporal dependencies of the channel and cross-dimensional dependencies between the environment and the channel to jointly predict the CSI for multiple future time slots.The proposed channel prediction framework is evaluated against traditional LSTM and Transformer models,demonstrating its superior performance.