CsiTransformer: A Limited-sample 6G Channel State Information Feedback Model
Qian Xiang, Xiaodan Wang, Xuan Wu, Jie Lai, Jiaxing He, Yafei Song · 2023
In recent years, the channel state information (CSI) feedback model based on artificial intelligence has emerged as a significant research area for 6G pre-research. However, existing methods mainly rely on the data-driven capabilities of deep learning models, and insufficient attention has been given to the CSI feedback problem under limited-sample conditions. Therefore, this paper proposed CsiTransformer, an improved Transformer-based CSI feedback model incorporating data augmentation techniques to handle the limited-sample CSI feedback problem. Experiments on different cellular scenarios demonstrate that CsiTransformer achieves an approximately 30% improvement in squared generalized cosine similarity compared to the traditional CsiNet when there are only 1000 samples for each scenario. Moreover, it shows a 49% improvement over the baseline model. The proposed CSI feedback model also exhibits good generalization across different cellular scenarios.