Chapter 26: 3D Gaussian Splatting for RF

Learning Objectives

• Describe the 3DGS representation (position, covariance, opacity, features) and explain how differentiable rasterisation achieves real-time rendering
• Formulate RF-3DGS by replacing optical colour with dB-scale received power and adapting the alpha-compositing integral for coherent summation
• Explain how RFCanvas fuses visual priors from LiDAR and camera with few-shot RF measurements via tensorial RF fields and spherical harmonics
• Analyse RadarSplat for automotive radar point cloud synthesis and FMCW-aware Gaussian rendering
• Compare NeRF, 3DGS, and SDF representations for RF scene reconstruction in terms of speed, accuracy, training data, and scalability