Physics-Aware Conditional SetGAN for Spatially Consistent Multi-User TR 38.901 Channel Generation

2026-07-13

AI Summary

This work addresses the high computational cost and trade-off between spatial consistency and efficiency in conventional TR 38.901 channel models (e.g., Sionna) for multi-user channel generation. The authors propose the first geometrically conditioned SetGAN that integrates physical priors by decoupling large-scale path loss from small-scale fading and modeling the geometric conditional distribution in latent space. Principal component analysis is employed to compress small-scale fading components, while Wasserstein distance is used to evaluate distributional fidelity. In UMa/NLoS scenarios, the method achieves a Wasserstein distance of only 0.41 dB for received power distribution, a median spatial consistency deviation below 0.03, a 3.45× speedup in generation time, and a 6.15× reduction in CPU overhead.

Abstract

TR 38.901-based channel models such as Sionna are reliable, but generating many multi-user channel realizations remains expensive. This paper asks a practical question: can a trained generative model produce multi-user TR 38.901 channels faster than Sionna without losing the spatial correlations imposed by user geometry? To answer this question, we propose a physics-aware, geometry-conditioned SetGAN trained on Sionna reference data. The method separates large-scale received power from normalized small-scale fading, compresses the latter with principal component analysis, and learns the conditional channel distribution in a latent space while preserving geometry-dependent correlations. On the UMa/NLoS benchmark, the model keeps the received-power distributions close to the reference, with about 0.41 dB Wasserstein distance, and reproduces spatial-consistency profiles with mean deviations below 0.03 on median curves versus distance. In addition, it reduces elapsed generation time by a factor of 3.45 and CPU-total cost by a factor of 6.15 relative to Sionna under matched user positions in the fixed-position CPU-vs-CPU benchmark. These results show that a trained generative model can substantially accelerate TR 38.901 channel generation without breaking the spatial consistency needed to evaluate multi-user systems.

Problem

• TR 38.901
• multi-user channel generation
• spatial consistency
• channel modeling
• computational efficiency

Innovation

• Physics-Aware Generative Model
• Conditional SetGAN
• Spatial Consistency
• TR 38.901 Channel Generation
• Geometry-Conditioned Learning