1: \begin{abstract}

2: Radio map estimation (RME), also known as \textit{spectrum cartography}, aims to reconstruct the strength of radio interference across different domains (e.g., space and frequency) from sparsely sampled measurements. 

3: To tackle this typical inverse problem, state-of-the-art RME methods rely on handcrafted or data-driven structural information of radio maps.

4: However, the former often struggles to model complex radio frequency (RF) environments and the latter requires excessive training---making it hard to quickly adapt to {\it in situ} sensing tasks.

5: This work presents a spatio-spectral RME approach based on {\it plug-and-play} (PnP) denoising, a technique from computational imaging.

6: The idea is to leverage the observation that the denoising operations of signals like natural images and radio maps are similar---despite the nontrivial differences of the signals themselves.

7: Hence, sophisticated denoisers designed for or learned from natural images can be directly employed to assist RME, avoiding using

8: radio map data for training.

9: Unlike conventional PnP methods that operate directly in the data domain, 

10: the proposed method exploits the underlying physical structure of radio maps and proposes an ADMM algorithm that denoises in a latent domain. This design significantly improves computational efficiency and enhances noise robustness.

11: Theoretical aspects, e.g., recoverability of the complete radio map and convergence of the ADMM algorithm

12: are analyzed.

13: Synthetic and real data experiments are conducted to demonstrate the effectiveness of our approach.

14: \end{abstract}

15: