基於 Transformer 的 MIMO OFDM 系統的通道估計

Abstract

隨著高速無線網路的需求逐步增加，通道矩陣估計在提升網路傳輸效能方面具有重要意義。針對該需求，本研究集中探討 MIMO OFDM 系統的通道矩陣估計，並引入智慧反射面 (IRS) 技術，旨在通過調整無線訊號路徑來增強系統效率和克服障礙物的影響。面對日漸複雜的無線通訊場景，精確的通道估計面對更加嚴峻的挑戰，因此我們設計了一種基於 Transformer 的創新方法來優化。過往的通道矩陣估計, 都只關注時域上的通道矩陣，我們使用了 cross-attention 來同時捕捉時域與頻域上通道矩陣的特徵。再加上利用通道矩陣的空間相關性與實虛交錯特性，使用了不同種的 sparse-attention 方法，大幅降低了計算開銷，從而實現更高的運算效率。同時我們也使用了卷積層來取代全連接層, 這是因為通道矩陣具有空間相關性。實驗結果說明，我們的方法在傳輸速率和系統效能上明顯優於傳統技術，表明通道矩陣估計在提升未來無線通訊性能的潛在應用價值。

As the demand for rapid wireless communication networks continues to increase, channel matrix estimation has become increasingly critical for improving network transmission performance. This study focuses on channel matrix estimation for MIMO OFDM systems and introduces intelligent reflecting surface (IRS) technology to enhance system efficiency and overcome obstacles by adjusting wireless signal paths. In today’s increasingly complex wireless communication environments, accurate channel estimation presents greater challenges. In response, we introduce an innovative Transformer-based approach for optimizing channel estimation in IRS-assisted MIMO-OFDM systems. Unlike previous methods that primarily focus on time-domain channel matrices, we leverage cross-attention to capture characteristics of the channel matrix across both time and frequency domains. Additionally, by integrating the spatial correlation and complex interleaved characteristics of the channel matrix, we apply various sparse-attention methods to effectively reduce computational costs and achieve higher processing efficiency. We utilized convolutional layers in place of fully connected layers, as the spatial correlation in the channel matrix makes them more suitable. The experimental findings indicate that this approach significantly outperforms traditional techniques in terms of transmission rate and system performance, underscoring the potential application value of channel matrix estimation in enhancing the performance of future wireless networks.