SPA-MAE: A Physics-Guided CSI Foundation Model for Wireless Physical Layer

2026-05-19

AI Summary

This work addresses the limited cross-task generalization of existing wireless foundation models, which stems from the absence of explicit modeling of channel physical characteristics during pretraining. To overcome this limitation, the authors propose SPA-MAE, the first channel state information (CSI) foundation model that integrates explicit physical priors into its pretraining framework. Built upon the Masked Autoencoder (MAE) architecture, SPA-MAE introduces dual branches—parameter-aware and structure-aware—that guide the encoder to learn multipath channel parameters and the sparsity structure of CSI in transform domains, respectively. Evaluated across four wireless physical-layer tasks, SPA-MAE consistently outperforms current state-of-the-art models, achieving superior generalization with fewer parameters, particularly under low signal-to-noise ratio and small-sample conditions.

Abstract

Deep learning (DL) has been widely used in future 6G physical layer communications, but task-specific DL models are difficult to generalize across different physical layer tasks. Recently emerging wireless foundation models demonstrate strong generalization capability. However, existing methods mainly adapt pretrained language/vision models or rely on CSI reconstruction objectives for pretraining, with limited use of channel knowledge, and thus have limited performance. To address this limitation, we propose SPA-MAE, a physics-guided wireless foundation model by exploiting the adapted MAE backbone and channel knowledge. A physical prior module is developed to provide two complementary guidance signals in the pretraining stage. Specifically, the parameter-aware guidance branch extracts features from explicit multipath parameters and encourages the encoder output to align them, while the structure-aware guidance branch encourages the encoder to capture the sparse transformed-domain CSI structure obtained after a 2D FFT. After end-to-end learning, the MAE encoder will be retained for downstream tasks. Experiments on four wireless tasks show that SPA-MAE outperforms state-of-the-art CSI foundation models with smaller number of parameters, especially under low-SNR and limited-data conditions.

Problem

• wireless foundation model
• channel state information
• physical layer
• generalization
• pretraining

Innovation

• physics-guided foundation model
• channel state information (CSI)
• masked autoencoder (MAE)
• multipath parameter awareness
• sparse CSI structure