A Spatial-Specific Neural Network-Based OFDM Channel Estimation Under Time-Varying Channels
Huang Wei, Jun Wang, Xiaonan Chen, Qihang Peng, Yuxi Zha, Li Li · IEEE Wireless Communications Letters · 2023
Deep learning (DL) has exhibited notable effectiveness in channel estimation for orthogonal frequency division multiplexing (OFDM) systems. However, most existing neural networks have not explicitly accounted for the intercarrier interference (ICI) and fail to achieve desirable performance under high-mobility air-ground communication environments. To address this problem, a modified Involution-based Channel Estimation Network (InvoEsNet) is proposed in this letter, which is designed to explicitly cope with ICI. The InvoEsNet consists of a modified involution-based preprocessing subnetwork (InvoPreNet) to mitigate ICI, followed by a residual channel state information subnetwork (ResCSINet) to further refine the channel estimates. Compared with existing convolutional networks, the proposed modified involution-based network is more suitable for dealing with ICI, especially under fast time-varying channels. Moreover, different weights are assigned to the training loss with respect to different signal-to-noise ratios to improve the neural network’s generalization. Experimental results show that our proposed InvoEsNet outperforms state-of-the-art benchmarks.