Toward Robust Channel Estimation in 5G With Atrous Pyramid Attention Networks

Abstract

Accurate channel estimation is crucial for reliable communication in modern wireless networks like fifth-generation (5G) and beyond 5G. However, current advanced methods face critical challenges in estimating channel coefficients accurately under noisy conditions and maintaining efficient model complexity. To overcome these challenges, this work proposes APANet, a deep convolutional network specially designed by incorporating atrous pyramid modules and synthesis attention modules to capture both local relationships and long-range dependencies in channel response matrices. Accordingly, APANet improves learning efficiency of complex channel patterns to increase the accuracy of channel estimation. Simulation results based on a synthesis channel dataset reveal that APANet outperforms state-of-the-art methods in estimation accuracy across various channel configurations while maintaining competitive computational efficiency. These results demonstrate the robustness and suitability of APANet for practical wireless networks, thus showcasing its potential to revolutionize channel estimation techniques.