RF-PGS is a novel framework for fully-structured spatial wireless channel representation based on Planar Gaussian Splatting, enabling accurate and efficient modeling of full radio propagation paths. [2508.16849] RF-PGS: A Fully-Structured Radio Propagation Representation with Planar Gaussian Splatting
Code will be released upon paper acceptance.
RF-PGS_demo.mp4
- Raw demonstration video: Google Drive
- Measured MPC visualization: Based on field measurements conducted by NIST USA, details available in this paper.
- Ground truth: Simulation results generated with the Sionna simulator, serving as the training and test data source for RF-PGS.
As the successor of our previous work RF-3DGS, RF-PGS introduces several key enhancements:
-
Greatly simplified training requirements:
The RF training stage now requires only the Rx-side path loss spectra while still enabling reconstruction of the time-of-flight (ToF) and angle-of-departure (AoD) for each multipath component (MPC). -
Fully-structured radio radiance field:
RF-PGS explicitly incorporates propagation distance into the model, allowing direct querying at the Tx-side (AoD domain) without transformation from the Rx-side (AoA domain). -
RF-specific enhancements:
By introducing Planar Gaussians with RF-oriented modifications, RF-PGS achieves more accurate environment representation and enables the extraction of surface-level RF parameters.