Source-linked AI summary
UrbanEMF: City-Scale EMF Mapping over a Continuous Urban Area with Real-World Base-Station Deployment
Shuangning Li, Chenxin Luo, Shanshan Wang, Yarui Zhang, Paul Lagouanelle, Joe Wiart
TL;DR
Existing datasets often lack continuous city-scale coverage with real BS information, while public exposure maps are not configurable sources for both aggregate and transmitter-specific radio conditions. UrbanEMF constructs a Paris dataset using real BS deployments and ray tracing, with multiple receiver heights and complementary RSS and path loss maps. It provides a dataset for studies ranging from coverage and exposure analysis to transmitter-specific propagation and link-level investigation.
Problem
Many existing datasets use spatially separated 2D scenes without real BS information, and public exposure maps are not configurable research datasets for both aggregate and transmitter-specific propagation information.
Source: §I · p. 2Method
UrbanEMF uses real urban geometry and operational BS locations in a city-scale ray-tracing pipeline, generating maps at multiple receiver heights.
Source: §I · p. 2, B. Key Characteristics · p. 2, 1) Accumulated RSS Map · p. 3Results
The resulting dataset provides spatially continuous Paris coverage with realistic BS deployments, multiple receiver heights, and both accumulated RSS and path loss maps.
Source: B. Key Characteristics · p. 2, 1) Accumulated RSS Map · p. 4, 2) Path Loss Map · p. 4Takeaways & Limitations
The maps support studies from area-level coverage and exposure analysis to transmitter-specific propagation and link-level investigation.
Source: §IV · p. 5Future work will extend the dataset to additional cities, carrier frequencies, deployment conditions, temporal variations, and richer channel-related outputs.
Source: §IV · p. 5
Abstract
from arXiv · showhide
Electromagnetic Fields (EMF) mapping is essential for wireless propagation modeling and are fundamental to a wide range of applications, including spectrum awareness, network planning, and integrated sensing and communication (ISAC). However, existing datasets are often limited to spatially separated 2D scenes without real base station (BS) information. To address this gap, we present UrbanEMF, a city-scale EMF mapping dataset constructed from real world urban geometry and physical BS deployments using ray tracing. Unlike conventional single scene-based datasets, UrbanEMF preserves continuous city-scale urban coverage and realistic transmitter location information. It further extends traditional 2D maps by incorporating multiple receiver heights. In addition, both aggregated received signal strength (RSS) maps and path loss maps are provided to characterize complementary aspects of the radio environment. This work provides a realistic and flexible benchmark for developing and evaluating learning-based methods in large-scale urban wireless environments. The code for this work is available at: https://github.com/lemonstudy/UrbanEMF
I. INTRODUCTION
Existing radio-map datasets often use isolated scenes or predefined transmitter configurations, while measurement datasets have collection constraints and public exposure maps are not configurable research datasets for both aggregate and transmitter-specific information. UrbanEMF addresses these gaps with a city-scale Paris dataset built around real BS deployments, multiple receiver heights, and complementary RSS and path loss maps.
I. INTRODUCTION
- Most existing simulation datasets consist of individual scenes or predefined transmitter configurations, rather than continuous real-city coverage with physical BS deployments.Measurement datasets are constrained by routes, density, observation time, and cost; public exposure maps are designed for monitoring rather than configurable research use.Source: §I · pp. 1–2
- UrbanEMF provides a city-scale Paris dataset using real BS locations, antenna tilt, multiple receiver heights, and both accumulated RSS and path loss maps.The framework can be extended to other cities or regions by replacing the urban environment and BS deployment data.Source: §I · p. 2
- UrbanEMF's accumulated RSS maps support area-level coverage and exposure analysis, while its per-BS path loss maps support transmitter-specific propagation studies.The two map types characterize received signal conditions and propagation attenuation, respectively.Source: §I · pp. 1–2
II. THE URBANEMF DATASET
The formulation represents a region covered by K base stations on a uniform H × W grid, with environmental and deployment information encoded as a multi-channel tensor. It defines aggregate received power and per-base-station path loss as complementary map outputs.
A. Problem Formulation
- The region is divided into an H × W grid, and its environment and BS deployment are represented as a C × H × W multi-channel tensor.Environmental channels describe propagation-relevant properties such as obstacle occupancy and building geometry.Source: A. Problem Formulation · p. 2
- Accumulated RSS is generated with Sionna ray tracing by computing total received power from nearby BSs and converting watts to dBm.The simulation accounts for line-of-sight propagation, specular reflection, and diffraction from the surrounding environment.Source: A. Problem Formulation · p. 2
- For each BS, the path loss map gives large-scale attenuation between that transmitter and receiver locations, derived from transmit power minus received power.Received power comes from propagation paths, so the resulting map reflects attenuation caused by distance.Source: A. Problem Formulation · p. 2
B. Key Characteristics
UrbanEMF combines operational BS locations and continuous Paris coverage with maps at multiple receiver heights. Its accumulated RSS and per-BS path loss outputs support distinct area-level and link-level studies.
B. Key Characteristics
- UrbanEMF uses operational BS locations from Cartoradio, preserving the relationship between actual urban geometry and physical BS deployment.The dataset also considers antenna tilt in its propagation modeling.Source: B. Key Characteristics · p. 2
- Paris is modeled as a spatially continuous environment, simulated district by district with surrounding areas rather than as independent urban scenes.The design preserves large-scale continuity and represents radio environments closer to practical dense urban networks.Source: B. Key Characteristics · p. 2
- Maps at multiple receiver heights provide propagation results at different vertical positions for the same horizontal location.These height-dependent maps support three-dimensional and height-aware urban propagation studies.Source: B. Key Characteristics · p. 2
- Accumulated RSS maps describe spatial received power for coverage and exposure analyses, while per-BS path loss maps describe attenuation for link-level studies.The paper compares UrbanEMF with existing radio-map and wireless-channel datasets; the accompanying prose describes their differing settings and transmitter configurations.Source: B. Key Characteristics · pp. 2–3, Table I · p. 3
III. GENERATION PROCESS
The dataset-generation process comprises real-world data acquisition, urban-scene construction, and Sionna-based ray-tracing simulation. The section details these steps and presents simulation results.
III. GENERATION PROCESS
- The generation process begins with real-world data acquisition.The passage identifies this as one of the three components of dataset generation.Source: §III · p. 3
- Urban-scene construction forms a distinct stage of the dataset-generation process.It is listed alongside real-world data acquisition and ray-tracing simulation.Source: §III · p. 3
- Sionna-based ray-tracing simulation is the third named component of dataset generation.The section also presents simulation results.Source: §III · p. 3
A. Data Acquisition and Scene Construction
UrbanEMF constructs city scenes from OSM building data and real BS records, then prepares them for ray-tracing simulation. The scene and transmitter setup includes explicit assumptions and memory-management steps.
A. Data Acquisition and Scene Construction
- OSM building footprints are projected to Lambert-93, extended 50 m beyond district boundaries, converted to 3D meshes, and exported as Sionna RT scenes.The boundary extension preserves neighboring geometry in simulations near district edges.Source: A. Data Acquisition and Scene… · p. 3
- Building heights use OSM height values or level-based estimates with 3 m uncertainty; missing values default to 15 m.Buildings and ground use concrete and medium-dry-ground materials, and the scene assumes flat terrain.Source: A. Data Acquisition and Scene… · p. 3
- Transmitter height uses Cartoradio’s height-above-ground record, while the transmitter is a cross-polarized dipole and the receiver is an isotropic vertical antenna.The simulated transmitter vertical coordinate is relative to local ground level rather than rooftop height.Source: A. Data Acquisition and Scene… · p. 3
- Transmitters are processed in batches and receiver subregions are bounded to reduce peak memory, with received powers aggregated before subregions are stitched.Each subregion still accounts for all relevant transmitters through sequential batch computation.Source: A. Data Acquisition and Scene… · p. 3
B. Sionna-Based EMF Map Simulation
The simulation section generates two complementary EMF map representations: exposure maps and path loss maps.
B. Sionna-Based EMF Map Simulation
- The section covers generation of both exposure maps and path loss maps as complementary EMF representations.The supplied passage identifies these representations but does not specify their generation procedures.Source: B. Sionna-Based EMF Map Simulation · p. 3
1) Accumulated RSS Map:
Accumulated RSS maps combine contributions from all relevant BSs on a 1 m grid. Their spatial patterns vary with receiver height, from building-shaped shadowing at low heights toward smoother coverage at higher positions.
1) Accumulated RSS Map:
- Each district’s accumulated RSS map evaluates received power on a 1 m regular grid while jointly considering all relevant BSs in the scene.Each grid value represents the aggregate contribution of those BSs.Source: 1) Accumulated RSS Map · p. 3
- At low receiver heights, accumulated RSS shows street- and building-shaped shadows; higher positions smooth the map as rooftop blockage gives way to large-scale transmitter geometry.The height-dependent pattern is shown for the same urban region.Source: Figure 1 · p. 4, 1) Accumulated RSS Map · p. 4
2) Path Loss Map:
Path loss maps are generated independently for individual BSs over local receiver regions. Their patterns reflect both Tx–Rx distance and surrounding building geometry.
2) Path Loss Map:
- For each BS, the path loss pipeline uses an 800 × 800 m receiver region centered on the transmitter at 5 m spatial resolution.Near district boundaries, the region extends into neighboring districts to preserve surrounding geometry and avoid truncation.Source: 2) Path Loss Map · p. 4
- PathSolver resolves multipath components for each Tx–Rx link and computes link gain from the sum of the resolved complex path coefficients.Unlike an aggregate RSS map, the path loss region does not include all Paris-area transmitters at every receiver.Source: 2) Path Loss Map · p. 4
- Dense scenes produce more fragmented path loss patterns, while open scenes show clearer distance-dependent attenuation and more extended low-loss regions.The contrast is attributed to frequent obstruction and multipath interactions in the denser scene.Source: 2) Path Loss Map · p. 4
3) Impact of Antenna Tilt Angle:
The section examines how antenna downtilt affects propagation using a directional 3GPP antenna pattern. With a 65° beamwidth, 12° downtilt leaves overall coverage similar while producing moderate local received-power variations.
3) Impact of Antenna Tilt Angle:
- Tilt angle α is measured between the antenna main-beam direction and the horizontal plane, with positive values denoting downward tilt.For a given BS and receiver height, the section defines the distance to where the main beam reaches the receiver plane.Source: 3) Impact of Antenna Tilt Angle · pp. 4–5
- The comparison uses a directional 3GPP TR 38.901 antenna pattern with no tilt and 12° downtilt.The directional pattern is used because antenna-orientation changes appear more clearly in its spatial antenna-gain distribution than with the preceding dipole pattern.Source: Figure 3 · p. 5, 3) Impact of Antenna Tilt Angle · p. 5
- Overall propagation structures and coverage remain highly similar with 12° downtilt, with moderate local received-power variations.The broad 65° half-power beamwidth leaves most receiver locations within or near the main radiation region.Source: 3) Impact of Antenna Tilt Angle · p. 5, Figure 3 · p. 5
IV. CONCLUSION
UrbanEMF provides continuous Paris-area EMF maps using real-world base-station deployments, with RSS and path-loss maps that retain multi-base-station and receiver-height information. The paper identifies expansion to additional cities and deployment conditions, fuller temporal variation, and richer channel outputs as future work.
IV. CONCLUSION
- UrbanEMF combines continuous Paris-area coverage and real-world BS deployments with RSS and path-loss maps, multi-BS information, and receiver-height data.The data are intended to support studies from area-level coverage and exposure analysis to transmitter-specific propagation and link-level investigation.Source: §IV · p. 5
- Future work will extend the pipeline to other cities, carrier frequencies, and deployment conditions, and investigate temporal variation and richer channel-related outputs.These extensions aim to broaden urban environments and improve dataset diversity and applicability.Source: §IV · p. 5