WiMAE (Wireless Masked Autoencoder)
收藏资源简介:
WiMAE是一个基于Transformer的编码器-解码器基础模型,在真实的多天线无线信道数据集上进行预训练。该数据集由加州大学欧文分校普遍通信与计算中心创建,旨在解决无线信道建模中的任务特定性问题。数据集包含大量的无线信道数据,用于训练WiMAE模型。WiMAE通过掩码重建的方式学习无线信道的结构特征,并通过对比学习增强模型的判别能力。该数据集在多个下游任务中表现出优异的性能,包括跨频段波束选择和视距检测。
WiMAE is a Transformer-based encoder-decoder foundation model pre-trained on real-world multi-antenna wireless channel datasets. Developed by the Center for Pervasive Communications and Computing at the University of California, Irvine, this dataset is designed to address the task-specificity issue in wireless channel modeling. It comprises a large corpus of wireless channel data for training the WiMAE model. WiMAE learns the structural features of wireless channels via masked reconstruction, and enhances the model's discriminative capability through contrastive learning. This dataset has demonstrated outstanding performance across multiple downstream tasks, including cross-band beam selection and line-of-sight (LoS) detection.




