Radio Frequency Radiance Modelling for Network Digital Twins [Ericsson]

The advent of next-generation wireless networks and the development of Network Digital Twins (NDT) have created an urgent need for accurate modeling of radio wave propagation in complex physical environments. Central to this modeling is to estimate how radio signals interact with surfaces through phenomena such as reflection, refraction, absorption, and transmission. Traditionally, the characterization of material Radio Frequency (RF) properties has been conducted in controlled laboratory settings using specialized equipment. However, as wireless networks become increasingly ubiquitous and applications such as Integrated Sensing and Communication (ISAC) emerge, there is growing interest in estimating surface RF properties directly within the physical environments where networks are deployed. This shift from laboratory to in-situ measurement presents significant technical challenges but offers the potential to enable scalable deployment of NDT technologies that can adapt to real-world conditions.

Existing approaches for learning RF properties in the physical world [1, 2] have shortcomings that hinder learning and deploying network digital twins at scale. Therefore, the goal of this project is to study in-depth and evaluate methods for surface RF properties estimation utilizing Radiance Fields techniques [3, 4].

This thesis project is sponsored by Ericsson Research and will be jointly supervised by Örebro University and Ericsson Research in Stockholm. Deliverables expected at the end of the thesis project include clearly documented working code, a final thesis report, and a live demonstration to an Ericsson audience.

Project objectives

1. Study and train RF Neural and Gaussian Splatting Radiance Fields models with data provided by Ericsson Research,
2. study and apply inverse rendering concepts [5, 6],
3. compare estimated surface RF properties against ground truth provided by Ericsson Research.

Who should apply

This project is suitable for students with strong programming skills in Python and C++ and hands-on experience training neural networks. You should be comfortable tackling challenging problems at the intersection computer vision and wireless communication, and motivated to work with real-world use cases.

Required qualifications: good programming skills in Python and C++, and hands-on experience with training neural networks.

Recommended qualifications include knowledge of signal processing and physics-based rendering for computer graphics.

[1] U. Virk et al., “On-Site Permittivity Estimation at 60GHz through Reflecting Surface Identification in the Point Cloud,” IEEE Transactions on Antennas and Propagation, vol. 66, no. 7, pp. 3599-3609, 2018.

[2] J. Hoydis et al., “Learning Radio Environments by Differentiable Ray Tracing,” IEEE Transactions on Machine Learning in Communications and Networking, vol 2, pp. 1527-1539, 2024.

[3] X. Zhao et al., “Nerf2: Neural radio-frequency radiance fields,” in Proc. the 29th Annual International Conference on Mobile Computing and Networking, pp. 1–15, 2023.

[4] Yang, et al., “GSRF: Complex-Valued 3D Gaussian Splatting for Efficient Radio-Frequency Data Synthesis,” arXiv 2025.

[5] H. Chen et al., “GI-GS: Global Illumination Decomposition on Gaussian Splatting for Inverse Rendering,” in Proc. ICLR, 2025.

[6] Z. Wu, et al., “3D Gaussian Inverse Rendering with Approximated Global Illumination,” arXiv, 2025.