Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Channel Modeling for FR3 Upper Mid-band via Generative Adversarial Networks

Yaqi Hu, Mingsheng Yin, Marco Mezzavilla, Hao Guo, Sundeep Rangan

TL;DR

This work tackles the need for data-driven, multi-frequency FR3 channel models by proposing a two-stage GAN framework: a link-state predictor conditioned on the link vector ${\mathbf{u}}$ and a conditional CWGAN-GP path generator that outputs the path vector ${\mathbf{x}}$ given $(\mathbf{u},s)$. It models the FR3 channel across $M=4$ frequencies ($6,12,18,24$ GHz) with up to $L=20$ shared paths per link, trained on urban NYC ray-tracing data. Results show accurate reproduction of both marginal and joint path loss statistics, SNR-beamforming differences, and RMS angular spreads when compared to ray-tracing ground truth, demonstrating the method’s capability to capture FR3 cross-frequency behavior. The approach offers a scalable, data-driven FR3 channel model suitable for cross-frequency network design and optimization in next-generation wireless systems.

Abstract

The upper mid-band (FR3) has been recently attracting interest for new generation of mobile networks, as it provides a promising balance between spectrum availability and coverage, which are inherent limitations of the sub 6GHz and millimeter wave bands, respectively. In order to efficiently design and optimize the network, channel modeling plays a key role since FR3 systems are expected to operate at multiple frequency bands. Data-driven methods, especially generative adversarial networks (GANs), can capture the intricate relationships among data samples, and provide an appropriate tool for FR3 channel modeling. In this work, we present the architecture, link state model, and path generative network of GAN-based FR3 channel modeling. The comparison of our model greatly matches the ray-tracing simulated data.

Channel Modeling for FR3 Upper Mid-band via Generative Adversarial Networks

TL;DR

This work tackles the need for data-driven, multi-frequency FR3 channel models by proposing a two-stage GAN framework: a link-state predictor conditioned on the link vector and a conditional CWGAN-GP path generator that outputs the path vector given . It models the FR3 channel across frequencies ( GHz) with up to shared paths per link, trained on urban NYC ray-tracing data. Results show accurate reproduction of both marginal and joint path loss statistics, SNR-beamforming differences, and RMS angular spreads when compared to ray-tracing ground truth, demonstrating the method’s capability to capture FR3 cross-frequency behavior. The approach offers a scalable, data-driven FR3 channel model suitable for cross-frequency network design and optimization in next-generation wireless systems.

Abstract

The upper mid-band (FR3) has been recently attracting interest for new generation of mobile networks, as it provides a promising balance between spectrum availability and coverage, which are inherent limitations of the sub 6GHz and millimeter wave bands, respectively. In order to efficiently design and optimize the network, channel modeling plays a key role since FR3 systems are expected to operate at multiple frequency bands. Data-driven methods, especially generative adversarial networks (GANs), can capture the intricate relationships among data samples, and provide an appropriate tool for FR3 channel modeling. In this work, we present the architecture, link state model, and path generative network of GAN-based FR3 channel modeling. The comparison of our model greatly matches the ray-tracing simulated data.
Paper Structure (12 sections, 7 equations, 6 figures, 3 tables)

This paper contains 12 sections, 7 equations, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. System Model
  3. Multi-Frequency Channel Model for FR3
  4. Generative Neural Network Modeling
  5. Link State Model
  6. Path Generative Network
  7. Data Acquisition: Urban FR3 Cellular System with Ray Tracing
  8. Modeling Results
  9. Path Loss with Omnidirectional Antennas
  10. SNR with Beamforming
  11. Root Mean Square (RMS) Angles Spread
  12. Conclusion

Figures (6)

  • Figure 1: The architecture for the proposed two-stage generative neural network model.
  • Figure 2: Ray tracing simulation area representing a 1100 $\times$ 570 $\text{m}^2$ urban region of NYC.
  • Figure 3: The CDF of the path loss for links within the test dataset is compared against the CDF of the pass loss for links randomly generated by the trained model.
  • Figure 4: The KDE plot of bivariate distributions of the path loss at 6GHz, 12GHz, 18GHz, and 24GHz.
  • Figure 5: The CDF plot of SNR differences (in dB) between beam selected by higher frequencies (12GHz, 18GHz, and 24GHz) and beam selected by 6GHz.
  • ...and 1 more figures