采用通道注意力机制的RIS辅助MIMO级联信道估计

可重构智慧表面（RIS）辅助多天线毫米波无线通信系统的级联信道有直传和反射两条链路,系统维数大且复杂,获取信道状态信息（CSI）困难. 针对RIS辅助下的大规模多输入多输出（MIMO）通信系统应用场景,提出了基于通道注意力机制的SE-ResNetV2网络来获取CSI,在残差网络（ResNet）中引入了通道注意力模块,通过挤压和激励策略来提高噪声特征的捕捉能力. 仿真结果表明,相比于最小二乘（LS）估计算法和常规的注意力机制深度ResNet,所提出的深度学习网络具有更好的去噪效果和估计精度.

Channel estimation for RIS assisted MIMO cascaded channels via attention mechanism

The cascaded channel of the reconfigurable intelligent surface （RIS） assisted multi-antenna millimeter-wave communication system had two links, the direct one and reflection one. The channel dimension was large and received noise feature was unpredictable which was a great challenge to obtain the channel state information （CSI） for application scenarios for massive multiple input multiple output （MIMO） communication system. An SE-ResNetV2 network based on channel attention mechanism was proposed to obtain CSI. SE-Net modules was introduced into the residual network （ResNet） to capture noise features through squeezing and excitation strategies. The simulation results showed that, the proposed deep learning network had better denoising effect and estimation accuracy compared to the least squares（LS） estimation algorithm and the conventional attention-mechanism deep ResNet.