Abstract: To address the problems of high pilot overhead and low accuracy in channel estimation for Simultaneously Transmitting and Reflecting Intelligent Surface (STAR-RIS) assisted mixed-field communication systems, a Multi-scale Feature-Aware Fusion Super Resolution Network (MFSRN) was proposed. Channel estimation problem was modeled as a matrix reconstruction task, firstly, common features were extracted from an upsampled channel matrix. Then, an attention gate was utilized to separate user-specific channel features, finally, reconstruction of multi-user channel matrices was completed in parallel. To enhance network performance, an Enhanced Convolution Block (ECB) and a Dual-Stream Feature-Aware (DSFA) module were designed to effectively extract both local and global features of mixed-field channel. Furthermore, a Feature Pyramid Attention (FPA) module was introduced to strengthen model's capability for multi-scale feature aggregation. Experimental results demonstrate that compared with classic DRSN and U-MLP schemes, proposed scheme achieves a lower pilot overhead while Normalized Mean Square Error (NMSE) is reduced by an average of 3.62 dB and 1.28 dB respectively at different Signal-to-Noise Ratios (SNR), shows better estimation performance.

摘要： 针对同时透射与反射智能表面（STAR-RIS）辅助混合场通信系统在信道估计中面临的导频开销高、精度低的问题，提出一种基于多尺度特征感知融合的超分重建信道估计网络（MFSRN）。该网络将信道估计问题建模为信道矩阵的重建任务，首先从上采样的信道矩阵中提取公共特征，随后利用注意力门限分离各用户特定信道特征，最终并行地完成多用户信道矩阵的重建。为提升网络性能，设计了增强卷积模块（ECB）和双流特征感知模块（DSFA）以有效提取混合场信道的局部和全局特征，并引入特征金字塔注意力模块（FPA）以增强模型对多尺度特征的聚合能力。实验结果表明，所提方案在实现更低导频开销同时，与经典的DRSN和U-MLP方案相比，归一化均方误差（NMSE）在不同信噪比（SNR）下平均降低了3.62 dB和1.28 dB，表现出更好的估计性能。