Effective outdoor pathloss prediction: A multi-layer segmentation approach with weighting map

TL;DR

The paper tackles outdoor path loss prediction by addressing the limitations of traditional methods through a deep learning framework that incorporates explicit propagation priors. It introduces multi-layer environmental inputs—Tx depth, Rx depth, distance maps—and a weighting map that prioritizes the direct Tx–Rx path, all processed by a ResNet-based backbone with a regression head. The approach yields consistent performance gains over PPNet, RPNet, and ViT, achieving reductions in RMSE by 1.2–3.0 dB across ITU AI/ML in 5G Challenge and ICASSP 2023 datasets while reducing FLOPs by about 60%. Ablation studies corroborate the importance of the distance map and, especially at higher frequencies, the weighting map for enhanced accuracy. These results suggest a practical, scalable path toward more accurate and efficient outdoor path loss prediction for next-generation wireless networks.

Abstract

Predicting pathloss by considering the physical environment is crucial for effective wireless network planning. Traditional methods, such as ray tracing and model-based approaches, often face challenges due to high computational complexity and discrepancies between models and real-world environments. In contrast, deep learning has emerged as a promising alternative, offering accurate path loss predictions with reduced computational complexity. In our research, we introduce a ResNet-based model designed to enhance path loss prediction. We employ innovative techniques to capture key features of the environment by generating transmission (Tx) and reception (Rx) depth maps, as well as a distance map from the geographic data. Recognizing the significant attenuation caused by signal reflection and diffraction, particularly at high frequencies, we have developed a weighting map that emphasizes the areas adjacent to the direct path between Tx and Rx for path loss prediction. {Extensive simulations demonstrate that our model outperforms PPNet, RPNet, and Vision Transformer (ViT) by 1.2-3.0 dB using dataset of ITU challenge 2024 and ICASSP 2023. In addition, the floating point operations (FLOPs) of the proposed model is 60\% less than those of benchmarks.} Additionally, ablation studies confirm that the inclusion of the weighting map significantly enhances prediction performance.

Figures (3)

• Figure 1: Illustration of the path distribution (red lines) between Tx (red point) and Rx (green point) using RT model, where only the rays with relatively high power at the Rx are kept.
• Figure 2: Overview of the path loss prediction architecture. In the data preprocessing stage, the environmental features are decomposed into the Tx-depth map, Rx-depth map, and distance map, followed by a Context Regional Cropping procedure. This procedure extracts a square region that encloses the Tx–Rx area together with its surrounding context, which serves as the reference region for subsequent feature extraction. In addition, a weighting map is introduced to assign higher importance to areas located near the direct propagation path between the Tx and Rx. The proposed model consists of a ResNet-based backbone for multi-level feature extraction and a regression head that estimates the path loss.
• Figure 3: Convergence analysis of the proposed model. Both training and validation loss curves indicate fast and stable convergence.