FiReT: A Neural Radiance Fields Framework for Wireless Field Reconstruction and Transmitter Placement

Negar Pouya, Armin Soleymani, Gholamreza Moradi, Farzaneh Abdollahi
Amirkabir University of Technology (Tehran Polytechnic)
  Paper   TechRxiv   Presentation Video

Accepted by IEEE IKT 2025

This video illustrates the impact of line-of-sight (LOS) and multipath propagation within a complex indoor environment. For a single transmitter and a representative set of receivers, it visualizes how rays are launched and propagate—directly and via reflections—to reach each receiver.

Abstract

We introduce FiReT, a deep learning framework designed for wireless channel estimation and transmitter placement optimization. Accurately modeling channel behavior in complex indoor spaces remains a critical challenge, as conventional approaches rely on exhaustive site surveys that demand dense measurements and significant human effort.

To overcome these limitations, our method leverages recent advances in Neural Radiance Fields (NeRF) to learn a continuous representation of the wireless radiation field from sparse observations. In our formulation, receivers are randomly positioned throughout the environment while the transmitter location is systematically varied, allowing the model to explore alternative deployment scenarios.

This design enables the estimation of wireless channels at unseen points and facilitates the identification of the globally optimal transmitter position that ensures robust coverage across the entire space. Through extensive evaluation, we show that our approach maintains high prediction accuracy with far fewer measurements than traditional methods, demonstrating the promise of NeRF-based learning in guiding efficient and scalable transmitter deployment for future wireless networks.

Model Architecture

FiReT learns a continuous wireless radiation field from sparse receiver measurements, supporting dense channel queries and transmitter ranking.
Residual-FiLM MLP injects DoA information across all layers via per-channel scaling and shifting to preserve angular sensitivity and stabilize training.

Channel Estimation & Transmitter Placement Results

FiReT achieves higher channel estimation accuracy with sparse samples compared to NeWRF and MLP baselines, reflecting effective NeRF-based field modeling.
Transmitter candidate ranking in the Bedroom scene. Candidates are placed on a (x, y, z) grid. “Average CFR |H|” is the mean CFR magnitude over all receivers and operating frequencies (unitless). “Ray Count” is the number of contributing propagation paths among all receivers after attenuation and phase screening. FiReT’s transmitter ranking balances link strength and spatial coverage using a weighted combination of CFR and ray-count metrics.

BibTeX

@inproceedings{pouya2025firet,
  title        = {FiReT: A Neural Radiance Fields Framework for Wireless Field Reconstruction and Transmitter Placement},
  author       = {Pouya, Negar and Soleymani, Armin and Moradi, Gholamreza and Abdollahi, Farzaneh},
  booktitle    = {Proceedings of the 16th International Conference on Information and Knowledge Technology (IKT)},
  year         = {2025}
}