physics-informed multipath mapping

This work addresses the limited accuracy of conventional radiomap modeling at high frequencies, which stems from neglecting fine-scale scatterers. The authors propose a multi-scatterer channel model based on spherical wave modal expansion that unifies the characterization of source radiation, single scattering, and multiple-scattering coupling effects through modal superposition. By reformulating the forward model as an inverse optimization problem, the approach jointly estimates scatterer responses and transmitter location. Notably, it integrates multi-scattering interactions with low-order modal approximation within a physically interpretable framework—a first in the field—and enables high-fidelity radiomap reconstruction and extrapolation from sparse measurements. Simulations demonstrate that the proposed model significantly outperforms existing methods in both spatial and beam domains, particularly in dense scattering environments.

Modeling rapidly time-varying wireless channels in dynamic scenarios—such as vehicle-to-vehicle (V2V) communications—remains challenging due to complex spatiotemporal propagation effects. Method: This paper systematically compares two mainstream approaches—differentiable ray tracing (DRT) and dynamic ray tracing (Dynamic RT)—and introduces the Multipath Lifetime Map (MPLM), a novel metric that jointly characterizes the spatiotemporal evolution of multipath components solely from static environmental geometry, thereby quantifying channel spatiotemporal coherence. Integrated within the 3DSCAT and Sionna simulation frameworks, the method is validated on a reproducible urban street-canyon scenario. Contribution/Results: Experimental results demonstrate strong agreement between MPLM predictions and measured channel data (mean correlation coefficient > 0.92), establishing MPLM as an interpretable, geometry-driven, and quantitative benchmark for evaluating and selecting dynamic propagation modeling techniques.

To address the poor generalizability and limited accuracy of conventional path loss models, this paper proposes a machine learning–based modeling approach that integrates multi-source GIS geospatial data with propagation geometry features. We innovatively construct an extended feature set encompassing terrain elevation, building height and material properties, street width, and three-dimensional propagation angles. Supervised learning algorithms—including XGBoost and Random Forest—are employed to develop a high-accuracy path loss prediction model. Evaluated on multi-region measurement datasets, the proposed model achieves a 32% reduction in mean absolute error compared to the Okumura-Hata and ITU-R P.1546 models. It demonstrates significantly improved cross-domain generalization across urban, suburban, and indoor environments while maintaining computational efficiency suitable for large-scale wireless network planning.

To address the challenge of constructing high-accuracy, high-efficiency radio maps (RMs) for 6G networks under multipath propagation, this paper identifies two key limitations: conventional electromagnetic (EM) simulations suffer from prohibitive computational cost and poor generalizability, while purely data-driven models neglect EM physics and fail to capture critical electromagnetic singularities. We establish, for the first time, a theoretical link between electromagnetic singularities and the negative-wavenumber regime of the Helmholtz equation. Building on this insight, we propose a dual-diffusion modeling framework: a singularity generation model that precisely captures multipath-induced singular features, and an environment fusion model that reconstructs full-field RMs under physical constraints imposed by the Helmholtz equation. Our approach jointly optimizes modeling fidelity and inference efficiency, achieving significant RM accuracy gains in complex, dynamic environments. It provides an interpretable, scalable paradigm for EM-aware environmental sensing in 6G.

This work addresses the challenge of efficiently constructing high-fidelity, physically consistent multipath radio frequency (RF) maps under sparse observations. It presents the first integration of physics-informed neural networks (PINNs) with graph neural networks (GNNs) to jointly model key multipath parameters—including path gain, time of arrival, and angle of arrival—in both 2D and 2.5D environments. By incorporating electromagnetic propagation priors and spatial correlations among receivers, the proposed approach ensures physical plausibility and structural coherence. A novel peak-weighted dynamic time warping metric is introduced to simultaneously account for amplitude errors and temporal alignment of dominant peaks. Experimental results demonstrate that the method significantly outperforms baseline approaches—including image-based models, diffusion models, and interpolation techniques—at both map-level and multipath-level evaluations, achieving strong cross-scenario generalization and high-accuracy sparse RF map completion.

This work addresses the high deployment cost and limited adaptability of traditional radio frequency (RF) propagation modeling, which typically relies on high-fidelity 3D maps or dense measurements and struggles in geospatially data-scarce environments. The paper introduces, for the first time, 3D Gaussian Splatting into map-free RF modeling, proposing an end-to-end learnable framework that reconstructs the propagation field using sparsely sampled RF measurements alone—without requiring prior knowledge of buildings or terrain. The approach employs anisotropic 3D Gaussian primitives initialized along transmitter–receiver paths and a learnable path-loss exponent. Evaluated on an outdoor sub-6 GHz dataset, the method achieves an RMSE of 5.38 dB, substantially outperforming existing techniques; in indoor BLE-based localization, it attains a mean error of merely 0.19 meters, improving upon NeRF² by nearly an order of magnitude.

This work addresses the limitations of existing wireless foundation models, which struggle to effectively model multipath effects due to their neglect of underlying propagation physics. To overcome this, the authors propose treating multipath propagation as the fundamental pretraining unit, representing transmitter–receiver links as continuous sequences of path tokens and predicting subsequent paths in an autoregressive manner. Built upon a Transformer architecture, the method innovatively integrates an environment-augmented retrieval (RAG) mechanism and a first-path codebook, enabling, for the first time, transferable path-level representations of wireless propagation. The approach significantly outperforms current models across multiple tasks—including localization, beam prediction, line-of-sight classification, and channel estimation—achieving a localization error as low as 5.57 meters and a Top-3 beam prediction accuracy of 0.914.

This work addresses the challenge of efficiently constructing and dynamically updating context-aware, location-dependent channel gain maps (CGMs) in large-scale wireless networks. It proposes a physics-informed approach based on 3D Gaussian Splatting (3DGS), introducing 3DGS for the first time into CGM modeling. The method represents the propagation environment using Gaussian primitives and integrates physical radio propagation mechanisms—including path loss, transmission, and scattering—to generate grid-level channel gains via differentiable rendering. To accommodate dynamic environmental changes, an incremental learning strategy is devised, combining frozen reference primitives with adaptable incremental ones. Experimental results demonstrate that the proposed method achieves high accuracy while significantly reducing computational overhead, enabling real-time updates and fine-grained representation of CGMs.

This work addresses the challenge of accurately modeling rapidly time-varying air-to-ground channels induced by the high-speed motion of low Earth orbit (LEO) satellites, particularly at low elevation angles where existing models fall short due to their neglect of real-world geographical effects such as terrain blockage and vegetation absorption. By integrating digital elevation models and land cover data with ray tracing to identify line-of-sight/non-line-of-sight links and reflection paths, the study incorporates diffraction, vegetation attenuation, and atmospheric loss models to rigorously quantify environmental impacts on signal propagation. Furthermore, it proposes a generative diffusion-based framework to learn the mapping from environmental context to channel characteristics, enabling, for the first time, physically accurate and scalable real-time channel prediction. Experiments demonstrate that the method significantly outperforms conventional statistical models on both cellular and LEO satellite measurement datasets, achieving precise channel state prediction for arbitrary satellite–ground geometries.