Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Efficient and Explainable KAN Framework forWireless Radiation Field Prediction

Jingzhou Shen, Xuyu Wang

TL;DR

This work tackles the challenge of accurate wireless channel modeling in environments with complex multipath and variability. It introduces KANNA-NeRF, a ray-in-ray-out encoder that combines a KAN-based attenuation and radiance network with masked Transformer attention to encode entire rays and capture spatial dependencies. Key contributions include an efficient PowerMLP-based KAN module, a dual-ray architecture with end-to-end differentiable ray tracing, ablation-informed design choices, and explainability visualizations. Empirical results on realistic and synthetic datasets show superior predictive accuracy (SNR/SSIM) over baselines, enabling more reliable wireless planning and localization with better generalization and interpretability.

Abstract

Modeling wireless channels accurately remains a challenge due to environmental variations and signal uncertainties. Recent neural networks can learn radio frequency~(RF) signal propagation patterns, but they process each voxel on the ray independently, without considering global context or environmental factors. Our paper presents a new approach that learns comprehensive representations of complete rays rather than individual points, capturing more detailed environmental features. We integrate a Kolmogorov-Arnold network (KAN) architecture with transformer modules to achieve better performance across realistic and synthetic scenes while maintaining computational efficiency. Our experimental results show that this approach outperforms existing methods in various scenarios. Ablation studies confirm that each component of our model contributes to its effectiveness. Additional experiments provide clear explanations for our model's performance.

An Efficient and Explainable KAN Framework forWireless Radiation Field Prediction

TL;DR

This work tackles the challenge of accurate wireless channel modeling in environments with complex multipath and variability. It introduces KANNA-NeRF, a ray-in-ray-out encoder that combines a KAN-based attenuation and radiance network with masked Transformer attention to encode entire rays and capture spatial dependencies. Key contributions include an efficient PowerMLP-based KAN module, a dual-ray architecture with end-to-end differentiable ray tracing, ablation-informed design choices, and explainability visualizations. Empirical results on realistic and synthetic datasets show superior predictive accuracy (SNR/SSIM) over baselines, enabling more reliable wireless planning and localization with better generalization and interpretability.

Abstract

Modeling wireless channels accurately remains a challenge due to environmental variations and signal uncertainties. Recent neural networks can learn radio frequency~(RF) signal propagation patterns, but they process each voxel on the ray independently, without considering global context or environmental factors. Our paper presents a new approach that learns comprehensive representations of complete rays rather than individual points, capturing more detailed environmental features. We integrate a Kolmogorov-Arnold network (KAN) architecture with transformer modules to achieve better performance across realistic and synthetic scenes while maintaining computational efficiency. Our experimental results show that this approach outperforms existing methods in various scenarios. Ablation studies confirm that each component of our model contributes to its effectiveness. Additional experiments provide clear explanations for our model's performance.
Paper Structure (23 sections, 16 equations, 9 figures, 2 tables)

This paper contains 23 sections, 16 equations, 9 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Preliminary
  3. Wireless Channel Prediction
  4. NeRF for Wireless Scenes
  5. KAN
  6. Framework
  7. CSI Data Pre-processing
  8. The Attenuation Network
  9. KAN Module
  10. Linear Transformer Module
  11. Feature Integration and Output
  12. The Radiance Network
  13. KAN Module
  14. Performer and Feature Integration
  15. Residual Connection and Output
  16. ...and 8 more sections

Figures (9)

  • Figure 1: Comparison between the traditional NeRF and our approach.
  • Figure 2: KANNA-NeRF framework based on NeRF$^2$.
  • Figure 3: Linear attention vs Traditional attention.
  • Figure 4: Similarity of mask attention and ray tracing.
  • Figure 5: CDF-SNR performance comparison across three datasets.
  • ...and 4 more figures