Research map / 3D vision / NeRF and radiance fields
NeWRF: A Deep Learning Framework for Wireless Radiation Field Reconstruction and Channel Prediction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes NeWRF, a NeRF-based framework that reconstructs wireless radiation fields from sparse channel measurements (with DoA guidance and a ray-searching algorithm) to predict channels at unvisited indoor locations.
Problem
Predicting wireless channels at unvisited locations in indoor environments from sparse channel measurements, to reduce the cost of site surveys; prior spatial predictors need dense measurements or analytical boundary setups.
Approach
Adapts NeRF volume rendering to wireless signals: backtraces rays along measured direction-of-arrivals, models a complex-valued radiated field with an MLP, sums contributions along paths with free-space propagation terms, and at test time estimates virtual transmitter locations via DoA ray-searching and voting.
Key results
- NeWRF outperforms KNN, MLP, and NeRF^2 baselines in all three simulated environments, staying above 5 dB SNR even in the largest office.
- Reaches high accuracy at 0.2 measurements/ft^3, about 800x lower density than NeRF^2 requires.
- Ray-searching for unknown DoA degrades SNR by only about 3-5 dB relative to ground-truth DoA.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use NeWRF instead of KNN, MLP, or NeRF² | Wireless channel prediction · Three simulated environments | SNR | Helped |
| reduce training measurement density to 0.2 measurements/ft^3 | NeWRF · Conference room | channel prediction accuracy | No clear effect |
| use the ray-searching algorithm to estimate DoA instead of ground-truth DoA | NeWRF · Unvisited locations | SNR: 3~5 dB | Hurt |
| increase environment size and complexity | NeWRF · Three simulated environments | SNR | Hurt |
| train using measurements with different additive noises across iterations instead of a fixed bias | NeWRF · Gaussian-noise robustness experiment | prediction accuracy | Helped |
| predict nearby subcarrier channels using a model trained on one subcarrier, without fine-tuning | NeWRF · Conference room subcarrier channel prediction | prediction accuracy | Helped |
Surprising findings
- The learned model places high volume density at virtual images of the transmitter behind reflective surfaces rather than on the reflectors themselves.
- NeRF^2 performs poorly at the low measurement density used here, since it requires about 178 measurements/ft^3.
Limitations
- Requires accurate phase measurements of narrowband channels, which are hard to obtain due to synchronization error.
- Requires DoA error within about one degree, needing very large antenna arrays.
- Performance degrades as environments grow larger and more complex; evaluation is only simulated.
Experimental setup
Datasets: MATLAB simulated channel datasets (conference room, bedroom, office) generated with Image Method ray tracing
Benchmarks: simulated conference room, bedroom, and office datasets
Metrics: SNR (-10 log10 NMSE) in dB
Baselines: K-Nearest Neighbors MLP NeRF^2
Builds on: NeRF NeRF^2 DBSCAN Image Method ray tracing
Closest papers by meaning
- Physics Unrolled Neural Operator for Wireless Field Modeling (NeurIPS 2026 · similarity 0.85)
- Can NeRFs "See" without Cameras? (NeurIPS 2025 · similarity 0.84)
- GSRF: Complex-Valued 3D Gaussian Splatting for Efficient Radio-Frequency Data Synthesis (NeurIPS 2025 · similarity 0.83)
- Differentiable and Learnable Wireless Simulation with Geometric Transformers (ICLR 2025 · similarity 0.83)
- Physics-Informed Generative Modeling of Wireless Channels (ICML 2025 · similarity 0.82)
- Radio Frequency Ray Tracing with Neural Object Representation for Enhanced RF Modeling (CVPR 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.