Enhancing UAV Networks Through Multi-Scale U-Net Deep Learning Framework

Research

NUST researcher develops a multi scale U-Net deep learning framework that cuts urban millimeter wave signal pathloss prediction error to 3.15 dB for reliable UAV network planning.

SDG 9 Industries, Innovation and Infrastructure
SDG 11 Sustainable Cities and Communities

As unmanned aerial vehicles (UAVs) become increasingly important to next-generation wireless networks, ensuring reliable communication in complex urban environments remains a significant challenge. Buildings, signal obstructions and varying flight altitudes can affect wireless signal propagation, making accurate coverage prediction essential for effective network planning.

UAV mmWave Coverage in 3D Urban Environment
Figure 1: 3D urban propagation modeling for UAV assisted millimeter-wave communication networks.

A New Deep Learning Approach for Pathloss Prediction

Addressing this challenge, Dr. Sajjad Hussain of the School of Electrical Engineering and Computer Science (SEECS), NUST, has developed a deep learning framework for predicting signal pathloss in UAV assisted millimeter wave communication networks. Published in IEEE Transactions on Wireless Communications (2026), the study introduces a multi scale feature extraction and fusion U Net architecture designed to generate more accurate radio coverage maps.

The proposed model combines information about transmitter distance, line of sight visibility and building locations to predict how wireless signals propagate across urban landscapes. It was evaluated using a purpose built dataset covering five urban locations including Munich, Helsinki, London and Manhattan at a carrier frequency of 28 GHz, as well as the publicly available RadioMapSeer benchmark.

Multi-Scale Feature Extraction and Fusion U-Net Pipeline
Figure 2: Comprehensive multi-scale feature extraction and fusion U-Net architectural pipeline for pathloss prediction.

Simulation & Evaluation Highlights

  • High Accuracy: Achieved a root mean square error (RMSE) of 3.15 dB on the in house UAV mmWave dataset and 3.97 dB on RadioMapSeer.
  • Robust Performance: Demonstrated consistent results across UAV altitudes of 25 to 45 meters.
  • Generalization: Proved effective when deployed across previously unseen urban environments.

By improving the accuracy and efficiency of radio propagation prediction, this research offers a promising approach to wireless network planning, particularly for UAV assisted communication systems in complex urban settings. The study also makes its source code and training pipeline publicly available to support reproducibility and further research.

Read the Paper

Sajjad Hussain, “A Multi Scale Feature Extraction and Fusion U Net for Pathloss Prediction in UAV Assisted mmWave Radio Networks,” IEEE Transactions on Wireless Communications, 2026.

About the Author

Dr. Sajjad Hussain

Assistant Professor | HOD, Department of Institute of Telecommunication (IOT-EE)

School of Electrical Engineering and Computer Science (SEECS), NUST

Leave a Reply

Your email address will not be published. Required fields are marked *