Doubly-Selective Channel Estimation Transformer Network with Varying Pilot Pattern

This paper presents a transformer-based doublyselective channel estimation (DSCE) network, conceptualized from the perspective of masked autoencoding, wherein pilot observations are regarded as the unmasked tokens and symbol observations as the masked tokens. To achieve robust channel estimation under varying pilot patterns, a relative positional encoding schema is adopted to incorporate the relative positions between tokens into the model, rendering the DSCE less sensitive to the absolute positions of pilot symbols. Additionally, the pilot augmentation strategy is implemented to simulate more diverse pilot patterns not presented in the training data, thereby mitigating overfitting and further improving the generalization ability of the model. The generalization and robustness of the proposed DSCE are validated through training and testing under various channel configurations. Comparative experiments demonstrate the superior performance of the proposed DSCE in terms of mean square error and bit error rate compared to traditional and deep learning-based methods, highlighting its state-of-the-art capabilities in channel estimation.