WiNeRF treats the antenna array's finite angular resolution and phase uncertainty as inductive constraints rather than assuming access to precise propagation geometry. The learned channel can be queried at previously unseen locations and used directly for downstream wireless tasks including beamforming, Angle-of-Arrival estimation, and coverage mapping.

Across approximately 30,000 synchronized CSI and pose measurements collected by a mobile robot in three indoor environments, WiNeRF achieved a 5.3 dB median prediction SNR and improved prediction SNR by approximately 3.0–6.7 dB over prior neural baselines.

I am currently designing a physics-grounded neural-operator model that learns the aggregate scattering behavior of vegetation from high-fidelity physical simulation. The goal is to replace expensive leaf- and branch-level interaction modeling with a compact learned representation that can eventually be integrated into wireless digital twins for fast and differentiable propagation modeling.

This project is currently in the architecture and model-design stage. Current work focuses on identifying the right physical abstraction and learning formulation for complex multi-interaction scattering inside vegetation.

I contributed to dataset generation and model development. The model achieved approximately 3.5 dB RMSE and 0.82 outage F1 against high-fidelity references. In one downstream access-point-placement experiment on a 10,000 m² floorplan, using the PU-HNO prediction resulted in approximately 92 Mbps throughput compared with 16 Mbps using a CNN prediction, despite relatively similar image-similarity scores.

WiViz uses an interactive 3D environment built with Unity and Sionna RT to expose mechanisms such as propagation, blockage, reflection, multipath, and interference. In an initial 24-participant study, users of the mechanism-revealing visualization correctly identified underlying wireless mechanisms 80.6% of the time versus 24.2% with an outcome-only representation.

I currently lead the design of the next WiViz user study, which focuses on wireless literacy, mechanistic reasoning, and transfer to unfamiliar scenarios, and I occasionally contribute to implementation.