Neural Representation for Wireless Radiation Field Reconstruction: A 3D Gaussian Splatting Approach

TL;DR

WRF-GS+, an enhanced framework that integrates electromagnetic wave physics into the neural network design, achieves state-of-the-art performance in the received signal strength indication (RSSI) and channel state information (CSI) prediction tasks, surpassing existing methods by more than 0.7 dB and 3.36 dB.

Abstract

Wireless channel modeling plays a pivotal role in designing, analyzing, and optimizing wireless communication systems. Nevertheless, developing an effective channel modeling approach has been a long-standing challenge. This issue has been escalated due to denser network deployment, larger antenna arrays, and broader bandwidth in next-generation networks. To address this challenge, we put forth WRF-GS, a novel framework for channel modeling based on wireless radiation field (WRF) reconstruction using 3D Gaussian splatting (3D-GS). WRF-GS employs 3D Gaussian primitives and neural networks to capture the interactions between the environment and radio signals, enabling efficient WRF reconstruction and visualization of the propagation characteristics. The reconstructed WRF can then be used to synthesize the spatial spectrum for comprehensive wireless channel characterization. While WRF-GS demonstrates remarkable effectiveness, it faces limitations in capturing high-frequency signal variations caused by complex multipath effects. To overcome these limitations, we propose WRF-GS+, an enhanced framework that integrates electromagnetic wave physics into the neural network design. WRF-GS+ leverages deformable 3D Gaussians to model both static and dynamic components of the WRF, significantly improving its ability to characterize signal variations. In addition, WRF-GS+ enhances the splatting process by simplifying the 3D-GS modeling process and improving computational efficiency. Experimental results demonstrate that both WRF-GS and WRF-GS+ outperform baselines for spatial spectrum synthesis, including ray tracing and other deep-learning approaches. Notably, WRF-GS+ achieves state-of-the-art performance in the received signal strength indication (RSSI) and channel state information (CSI) prediction tasks, surpassing existing methods by more than 0.7 dB and 3.36 dB, respectively.

Figures (15)

• Figure 1: Different types of wireless channel modeling.
• Figure 2: The synthesized spatial spectra of five algorithms in a laboratory environment. The spatial spectrum represents the signal strength received from different directions composed of the azimuthal and elevation angles. (a) shows the 3D point clouds of the laboratory created by LiDAR, which are used for the ray tracing algorithm. In this setting, the TX can be located at any position, while the RX is equipped with a $4 \times 4$ antenna array and fixed at a corner. (b) compares the synthesized spatial spectra of five algorithms when the TX is located at four different positions, i.e., P1-P4, as shown in (a). The ground truth is obtained using the antenna array.
• Figure 3: The AoA computation of the received signal with an antenna array. (a) There are $K$ antennas deployed in a grid, and the signal source is located at point B. (b) The spatial spectrum is generated from the antenna array. The illustration of the spatial spectrum in Fig. \ref{['SynSS']}(c) is obtained by projecting the spatial spectrum to the X-Y plane.
• Figure 4: An overview of the WRF-GS framework. The 3D points, which can be randomly distributed or captured via LiDAR sensors, and the position of the TX are first passed into a scenario representation network. This network represents the virtual TXs in the scene using a set of 3D Gaussians, each of which carries environmental attenuation information and signal characteristics. To project these 3D Gaussian representations onto the perception plane of the RX antenna array, the Mercator projection is employed. Finally, the resulting spatial spectra are rendered using the electromagnetic splatting method.
• Figure 5: Architecture of the neural model. The model comprises two MLPs. The first MLP processes the input 3D point clouds to capture the spatial attenuation information of the environment, which is independent of the specific TX position. This MLP outputs the attenuation and a feature vector for each spatial location. The second MLP then combines the spatial feature vector with the TX position to capture the signal characteristics.