![[2602.22231] FM-RME: Foundation Model Empowered Radio Map Estimation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.22231] FM-RME: Foundation Model Empowered Radio Map Estimation
Summary
The paper presents FM-RME, a foundation model for radio map estimation that integrates self-supervised learning and physical propagation knowledge to enhance zero-shot generalization in multi-dimensional spectrum environments.
Why It Matters
FM-RME addresses significant limitations in traditional radio map estimation methods by leveraging advanced machine learning techniques. This innovation is crucial for improving data efficiency and accuracy in complex wireless environments, making it relevant for fields such as telecommunications and AI-driven signal processing.
Key Takeaways
- FM-RME utilizes a geometry-aware feature extraction module for improved radio map estimation.
- The model supports zero-shot inference, eliminating the need for scenario-specific retraining.
- A masked self-supervised pre-training strategy enhances generalizability across diverse wireless environments.
- Simulation results demonstrate FM-RME's superior performance compared to existing methods.
- The integration of physical knowledge with data-driven approaches marks a significant advancement in radio map estimation.
Electrical Engineering and Systems Science > Signal Processing arXiv:2602.22231 (eess) [Submitted on 15 Feb 2026] Title:FM-RME: Foundation Model Empowered Radio Map Estimation Authors:Dong Yang, Yue Wang, Songyang Zhang, Yingshu Li, Zhipeng Cai, Zhi Tian View a PDF of the paper titled FM-RME: Foundation Model Empowered Radio Map Estimation, by Dong Yang and 5 other authors View PDF HTML (experimental) Abstract:Traditional radio map estimation (RME) techniques fail to capture multi-dimensional and dynamic characteristics of complex spectrum environments. Recent data-driven methods achieve accurate RME in spatial domain, but ignore physical prior knowledge of radio propagation, limiting data efficiency especially in multi-dimensional scenarios. To overcome such limitations, we propose a new foundation model, characterized by self-supervised pre-training on diverse data for zero-shot generalization, enabling multi-dimensional radio map estimation (FM-RME). Specifically, FM-RME builds an effective synergy of two core components: a geometry-aware feature extraction module that encodes physical propagation symmetries, i.e., translation and rotation invariance, as inductive bias, and an attention-based neural network that learns long-range correlations across the spatial-temporal-spectral domains. A masked self-supervised multi-dimensional pre-training strategy is further developed to learn generalizable spectrum representations across diverse wireless environments. Once pre-trai...
