面向毫米波大规模MIMO系统的信道估计OA

Massive multiple-input multiple-output(Massive MIMO)technology significantly enhances spectral efficiency and network capacity through large-scale antenna deployment and is recognized as one of the pivotal technologies for future wireless communications.However,the proliferation of antennas introduces prohibitively high pilot overhead and feedback overhead.To address these challenges,a low-overhead channel estimation algorithm is proposed,targeting the channel sparsity characteris-tics of time-division duplexing(TDD)millimeter wave(MMW)Massive MIMO systems.The algorithm is based on an improved learned approximate message passing(LAMP)neural network enhanced by convolutional neural networks(CNNs),where a re-sidual structure is integrated with the efficient channel attention(ECA)mechanism into the CNN architecture.By leveraging ECA's dynamic weight allocation to amplify critical channel features,the proposed CNN replaces the conventional nonlinear shrinkage function in the iterative unfolding layers of LAMP,thereby achieving adaptive enhancement of nonlinear optimization.Theoretical analysis and simulation results demonstrate that the proposed algorithm reduces the normalized mean square error(NMSE)by an average of 10 dB compared to compressed sensing(CS)algorithms,and outperforms deep learning-based LAMP and GM-LAMP algorithms by approximately 1.8 dB.Furthermore,it effectively reduces system overhead while significantly im-proving the accuracy of channel estimation.