Path loss prediction for air-to-ground communication links via scenario transfer technology

Guanjie Zhang, T. T. Lei, Xiaofeng Huo, Yanbin Li, Mengjie Geng, Hongjie Yang, Yuxin Yang, Xiaomin ChenComplex 2024 · DOI

Abstract

Path loss (PL) is a significant channel parameter for the link budget in unmanned aerial vehicle-aided communications. This study introduces an innovative neural network model to estimate PL for air-to-ground communication links. Utilizing the geometric characteristics of varied physical environments, the model accurately predicts PL in diverse communication scenarios. A back-propagation neural network technique is introduced for extrapolating PL under both line-of-sight and non-line-of-sight conditions. A dataset acquisition strategy, comprising scenario reconstruction and advanced ray-tracing techniques, is employed to foster the model’s training and evaluation. Finally, the proposed model is fully trained in diverse communication scenarios, and then used to predict the PLs in a new communication scenario generated by the International Telecommunication Union standard at 28 GHz. The results demonstrate that the extrapolated PLs of the proposed model are well consistent with the reference results. As existing PL models and standard PL models aim at several specifically defined scenarios, the proposed model can predict the PLs in some undefined and unknown scenarios.

Introduction

As unmanned aerial vehicles (UAVs) transition from recreational use to critical components of 5G/6G networks and emergency medical logistics, the "link budget"—the comprehensive accounting of all gains and losses in a communication system—becomes a high-stakes calculation. Central to this is Path Loss (PL): the reduction in power density of an electromagnetic wave as it propagates through space. Accurate PL prediction is essential for ensuring reliable connectivity, especially in complex urban environments where signals are easily obstructed.

Predicting signal strength for Air-to-Ground (A2G) links is notoriously difficult. In urban areas, buildings create a chaotic maze of reflections, diffractions, and "shadows." Traditionally, engineers have relied on two main approaches:

  1. Deterministic Models (e.g., Ray Tracing): These provide high accuracy but are computationally expensive and slow, requiring detailed 3D maps of every building in the environment.
  2. Empirical Models (e.g., 3GPP standards): These offer fast calculations based on generalized formulas but often lack the precision needed for specific or unusual city layouts, particularly at varying drone altitudes.

A recent study by Zhang et al., published in Complex Engineering Systems, addresses this gap by proposing a machine-learning approach that utilizes "scenario transfer" to estimate PL in environments the model has never encountered.

Methodology

The researchers developed a Back-Propagation Neural Network (BPNN) designed to predict millimeter-wave (28 GHz) path loss. The core innovation lies in the selection of input features. Rather than relying solely on distance and height, the model incorporates three critical geometric parameters defined by the International Telecommunication Union (ITU-R):

  • $α$: The ratio of land area covered by buildings to the total area.
  • $β$: The average number of buildings per square kilometer.
  • $γ$: A parameter determining the distribution of building heights, typically modeled using a Rayleigh distribution.

By integrating these statistical descriptors of urban geometry, the network learns the fundamental relationship between city structure and signal decay.

To generate a robust dataset for training and evaluation, the team employed a "Scenario Reconstruction" strategy. They built virtual cities and utilized advanced Ray Tracing simulations at 28 GHz to produce "ground truth" path loss values. This approach allowed for the creation of a massive, high-fidelity dataset:

  • Training/Testing Sets: Data derived from four representative scenarios (Suburban, Urban, Dense Urban, and Highrise) alongside 176 varied Non-Line-of-Sight (NLoS) configurations.
  • Scale: The final dataset comprised 6,078 points for Line-of-Sight (LoS) conditions and 49,230 points for NLoS conditions.
  • Validation: A completely novel scenario ($α=0.4, β=400, γ=18$) was used to test the model’s generalization capabilities through scenario transfer.

Results and Discussion

The BPNN model demonstrated exceptional precision when compared against the Ray Tracing reference data. For LoS conditions, the model achieved a Root Mean Square Error (RMSE) of only 0.2493 dB, with most deviations remaining under 3 dB. In the more challenging NLoS conditions, the model reached an RMSE of 8.27 dB. While naturally higher due to the complexity of signal blocking and scattering, 80% of the prediction errors were within 5 dB, representing a significant achievement for such environments.

Several key insights emerged from the study:

  • Geometry as a Primary Feature: By treating city statistics as primary inputs, the model moves beyond "memorizing" specific maps. It develops a generalized understanding of how urban density and building height distributions affect electromagnetic physics.
  • The Impact of Altitude: The study confirmed that for UAVs, horizontal distance has a more severe impact on signal loss at lower altitudes. As the UAV ascends, it clears obstacles, effectively lengthening the LoS path and reducing interference from ground-level "clutter."
  • Superiority over Standard Models: When benchmarked against 3GPP standard models, the BPNN was significantly more consistent with simulation results, particularly at the specific altitudes where UAVs typically operate.

Conclusion

The proposed BPNN model offers a powerful middle ground between the speed of empirical formulas and the accuracy of deterministic simulations. By leveraging scenario transfer technology, it can predict path loss in undefined and unknown urban environments with high reliability.

However, limitations remain. The model was trained on synthetic Ray Tracing data, and the researchers plan to validate it using real-world field measurements from a physical 28 GHz A2G channel sounder. Future work will also look to incorporate additional environmental features, such as terrain relief and foliage, to further enhance prediction accuracy.

Paper Citation: Zhang G, Lei T, Huo X, et al. Path loss prediction for air-to-ground communication links via scenario transfer technology. Complex Eng Syst 2024;4:18. DOI: 10.20517/ces.2024.55

Summary updated

Abstract

Path loss (PL) is a significant channel parameter for the link budget in unmanned aerial vehicle-aided communications. This study introduces an innovative neural network model to estimate PL for air-to-ground communication links. Utilizing the geometric characteristics of varied physical environments, the model accurately predicts PL in diverse communication scenarios. A back-propagation neural network technique is introduced for extrapolating PL under both line-of-sight and non-line-of-sight conditions. A dataset acquisition strategy, comprising scenario reconstruction and advanced ray-tracing techniques, is employed to foster the model’s training and evaluation. Finally, the proposed model is fully trained in diverse communication scenarios, and then used to predict the PLs in a new communication scenario generated by the International Telecommunication Union standard at 28 GHz. The results demonstrate that the extrapolated PLs of the proposed model are well consistent with the reference results. As existing PL models and standard PL models aim at several specifically defined scenarios, the proposed model can predict the PLs in some undefined and unknown scenarios.
BibTeX
@inproceedings{zhang2024path,
  author = {Guanjie Zhang and T. T. Lei and Xiaofeng Huo and Yanbin Li and Mengjie Geng and Hongjie Yang and Yuxin Yang and Xiaomin Chen},
  title = {{Path loss prediction for air-to-ground communication links via scenario transfer technology}},
  booktitle = {Complex},
  year = {2024},
  doi = {10.20517/ces.2024.55},
}
Figure