CE-TrEM: Lightweight OFDM channel estimation based on attention mechanism
Author Identifier (ORCID)
Abstract
Accurate channel estimation in orthogonal frequency-division multiplexing (OFDM) systems is crucial for supporting real-time applications in the fifth-generation (5G) mobile networks. Existing channel estimation methods often suffer from poor accuracy under low signal-to-noise ratio (SNR) conditions and involve excessive model complexity and computational overhead. In this paper, we propose CE-TrEM, a lightweight channel estimation model that integrates an attention-based architecture with a prior-free Stein's Unbiased Risk Estimate Adaptive Regularized Least Squares (SURE-ARLS) preprocessing mechanism. The model takes the preprocessed channel matrix and an explicitly constructed spatial confidence map as a three-channel input tensor, which is sequentially processed by a Transformer encoder and a lightweight residual network. The refined features are then up-sampled to reconstruct the full channel estimate. This architecture achieves a substantial reduction in model parameters while maintaining high accuracy. Simulations validate the effectiveness of the model across a wide variety of scenarios and a broad range of SNR conditions, highlighting especially its substantial accuracy gains under low SNRs. The model maintains a lightweight design and achieves up to an 81.14% reduction in error compared to baseline approaches.
Keywords
5G, attention mechanism, channel estimation, lightweight model
Document Type
Conference Proceeding
Date of Publication
1-1-2026
Publication Title
2026 9th International Conference on Communication Engineering and Technology (ICCET)
Publisher
IEEE
School
School of Engineering
ISBN
[9798319506900]
Copyright
subscription content
First Page
264
Last Page
269
Recommended Citation
Gao, Y., Chen, X., Hu, S., Ni, W., & Wang, X. (2026, June). CE-TrEM: Lightweight OFDM channel estimation based on attention mechanism. In 2026 9th International Conference on Communication Engineering and Technology (ICCET) (pp. 264-269). IEEE. https://doi.org/10.1109/ICCET70051.2026.11659042