# PropSplat: Map-Free RF Field Reconstruction via 3D Gaussian Propagation Splatting

Authors: William Bjorndahl, Maninder Pal Singh, Farhad Nouri, Joseph Camp.
Affiliations: William Bjorndahl, Farhad Nouri, and Joseph Camp — Southern Methodist University. Maninder Pal Singh — University of Houston.

Paper website: https://propsplat.com/
Published in: 2026 IEEE International Symposium on Spectrum Innovation (DySPAN), pages 217–224 (2026).
IEEE DOI: https://doi.org/10.1109/DySPAN69846.2026.11571104
arXiv: https://arxiv.org/abs/2605.08035
Author manuscript PDF (arXiv v1): https://propsplat.com/propsplat.pdf
BibTeX: https://propsplat.com/citation.bib

## Abstract

Building a site-specific propagation model typically requires either ray-tracing over detailed 3D maps or dense measurement campaigns. Both approaches are expensive and often infeasible for rapid deployments where geographic data is unavailable or outdated. We present PropSplat, a map-free propagation modeling method that reconstructs radio frequency (RF) fields using 3D anisotropic Gaussian primitives. Each Gaussian encodes a scalar path loss offset relative to an explicit baseline path loss model with a learnable path loss exponent. Gaussians are initialized along observed transmitter–receiver paths and optimized end-to-end to learn the propagation environment without external information like floor plans, terrain databases, or clutter data. We evaluate PropSplat against wireless radiance field methods NeRF², GSRF, and WRF-GS+ on two real-world datasets. On large-scale outdoor drive-tests spanning multiple topographical regions at six sub-6 GHz frequencies, PropSplat achieves 5.38 dB RMSE when training measurements are spaced 300 m apart and outperforms WRF-GS+ (5.87 dB), GSRF (7.46 dB), and NeRF² (14.76 dB). On indoor Bluetooth Low Energy measurements, PropSplat achieves 0.19 m mean localization error, an order of magnitude better than NeRF² (1.84 m), while achieving near-identical received signal strength prediction accuracy. These results show that accurate site-specific propagation reconstruction is achievable from sparse RF-native measurements. The need for geographic data as a prerequisite for scalable RF environment modeling is reduced.

## Method and related work

PropSplat studies map-free RF field reconstruction, related to wireless radiance fields, radio radiance fields, and radio Gaussian splatting. Inputs are transmitter and receiver positions, frequency, and measured path loss or RSSI. Each learnable 3D anisotropic Gaussian contributes an independent scalar path loss offset to a log-distance baseline with a learnable exponent. Gaussians encode empirical localized propagation corrections, not reconstructed buildings or other physical geometry.

Comparisons in the paper include NeRF² (https://doi.org/10.1145/3570361.3592527), GSRF (https://openreview.net/forum?id=E3oNDQ8e9r), and WRF-GS+ (https://doi.org/10.1109/TWC.2025.3631663).

## Outdoor RF field reconstruction

Ofcom drive-test measurements cover seven UK locations and six frequencies from 449 to 5850 MHz, spanning urban, suburban, rural, and mountainous environments. Training samples are spaced approximately 300 m apart, using less than 1% of the measurements.

![London drive-test reconstruction at 5850 MHz: sparse training measurements, reconstructed field, and held-out test errors.](https://propsplat.com/assets/propsplat-london-5850mhz.png)

Figure 2. London at 5850 MHz: 375 training measurements are selected from more than 130,000 drive-test points. [Vector figure (PDF)](https://propsplat.com/assets/propsplat-london-5850mhz.pdf).

Results: path loss RMSE averaged across all city–frequency pairs, using the same approximately 300 m spatial training split:

| Method | Path loss RMSE (dB) |
| --- | ---: |
| PropSplat | 5.38 |
| WRF-GS+ | 5.87 |
| GSRF | 7.46 |
| NeRF² | 14.76 |

Impact: accurate coverage reconstruction from sparse measurements could reduce drive-test effort and support crowdsourced network optimization, including locations where building maps and terrain data are unavailable.

## Indoor BLE signal modeling and localization

The public NeRF² Bluetooth Low Energy dataset contains measurements from 6,000 transmitter positions and 21 fixed gateways in a 15,000 sq. ft. nursing home. PropSplat predicts received signal strength (RSSI) and generates fingerprints for indoor positioning.

![Two example transmitter-to-gateway paths through learned Gaussians and cumulative predicted RSSI along each path.](https://propsplat.com/assets/propsplat-ble.png)

Figure 3. Each Gaussian's contribution can be traced along a propagation path. These two examples show opposite corrections to the baseline: 17.7 dB of added loss for TX₁ and a −22.3 dB correction for TX₂. The plots illustrate individual predictions; aggregate test results are reported below. [Vector figure (PDF)](https://propsplat.com/assets/propsplat-ble.pdf).

Results under identical data splits:

| Metric | Dense: 4,200 training positions (70%) | Sparse: 758 training positions (12.6%) |
| --- | ---: | ---: |
| PropSplat mean localization error | 0.19 m | 0.44 m |
| NeRF² mean localization error | 1.84 m | 7.33 m |
| PropSplat RSSI prediction RMSE | 4.74 dB | 6.51 dB |
| NeRF² RSSI prediction RMSE | 4.68 dB | 7.17 dB |

Impact: similar signal-strength accuracy can lead to very different positioning accuracy. PropSplat preserves useful differences between nearby locations, supporting precise indoor positioning even with sparse measurements and no floor plan.

The model targets static environments; temporal variation, frequency-selective fading, and multi-antenna effects remain future work.

## Figures

- [Propagation concept](https://propsplat.com/assets/propsplat-concept.png): Gaussian scalar corrections to baseline path loss. Buildings are illustrative and not inputs.
- [London 5850 MHz](https://propsplat.com/assets/propsplat-london-5850mhz.png): 375 sparse training measurements selected from more than 130,000 drive-test points.
- [Indoor BLE paths](https://propsplat.com/assets/propsplat-ble.png): two example propagation paths with individually traceable Gaussian contributions.

## Citation

```bibtex
@INPROCEEDINGS{11571104,
  author={Bjorndahl, William and Singh, Maninder Pal and Nouri, Farhad and Camp, Joseph},
  booktitle={2026 IEEE International Symposium on Spectrum Innovation (DySPAN)},
  title={PropSplat: Map-Free RF Field Reconstruction via 3D Gaussian Propagation Splatting},
  year={2026},
  volume={},
  number={},
  pages={217-224},
  keywords={Modeling;Measurement;Equations;Training;Neural radiance field;Propagation;Radio frequency;Location awareness;Testing;Received signal strength indicator;RF field reconstruction;3D Gaussian splatting;sparse measurements;path loss prediction;machine learning},
  doi={10.1109/DySPAN69846.2026.11571104}}
```