基于混合IRS辅助大规模MIMO系统的仿真信道估计方法

Simulation Channel Estimation Method for Large-Scale MIMO Systems Based on Hybrid IRS

对于IRS辅助的大规模MIMO系统,大多数研究都需要基于信道状态信息已知,而IRS通常为无源中继,导频开销较大,信道估计具有挑战性。为此,研究引入了一种包含有源和无源元件的混合IRS架构,使用少量RF链接收用户发送的上行导频信号,利用毫米波信道的稀疏特性,采用压缩感知算法重构信道,减少了导频损耗。考虑到信道为复数矩阵,传统的方法都将其实部虚部分开输入网络进行训练,该类方法会丢失信道的部分信息。为此,研究引入了一种注意力引导的复数深度去噪的神经网络AM-DnCNN。该网络可以将信道看作是二维带有噪声的矩阵进行训练,引入注意力机制加强信道的噪声特征,网络输出噪声矩阵,重构噪信道矩阵。仿真结果表明,所提方法可以利用更少的导频获得更优的信道状态信息,有效减少了导频损耗,且在不同路径数量和不同信噪比的情况下,网络也具有很好的鲁棒性。

For IRS-assisted large-scale MIMO systems, most studies need to be based on channel state infor-mation known, and IRS is usually passive relay, with high pilot overhead and challenging channel estimation. To this end, a hybrid IRS architecture containing active and passive components is in-troduced. A small number of RF links are used to receive upstream pilot signals sent by users. The sparse characteristics of millimeter wave channels are utilized to reconstruct the channels by com-pressed sensing algorithm to reduce pilot losses. Considering that the channel is a complex matrix, traditional methods separate the real and imaginary parts into the network for training, which will lose some information of the channel. Therefore, an attention-guided complex depth denoising neural network AM-DnCNN is introduced. In this network, the channel can be regarded as a two-dimensional matrix with noise for training. Attention mechanism is introduced to enhance the noise characteristics of the channel. The network outputs the noise matrix and reconstructs the de-noised channel matrix. The simulation results show that the proposed method can use fewer pi-lots to obtain better channel state information, effectively reduce pilot loss, and the network also has good robustness under different number of paths and different SNR.