Attention-Based Neural Network for OFDM Channel Estimation

Le résumé fourni par la source

This paper proposes an attention-based neural network for OFDM channel estimation, designed to refine conventional least-squares (LS) estimates by exploiting time-frequency correlation characteristics of wireless channels. The proposed network incorporates a lightweight attention mechanism with query, key, and value mappings, enabling adaptive symbol-wise filtering while maintaining low computational complexity suitable for online deployment. An online training framework is introduced to adapt the network to time-varying channels with limited training samples. Simulation results demonstrate that the proposed method significantly outperforms conventional LS and MMSE estimators in terms of mean square error (MSE), especially under low signal-to-noise ratio (SNR) conditions. Furthermore, over-the-air experiments using a USRP testbed validate the practical feasibility and robustness of the method in realistic multipath fading environments.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.