Deep Learning-Based Traffic Prediction for Non-Terrestrial Networks: A Hybrid Satellite-UAV Approach

View at publisher Summarise with VIRAT
Authors: Vaskar Chakma, Seong-Yong Moon, Wooyeol Choi
Cited by: 0
References: 12

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The rapid growth of mobile data and IoT devices demands seamless wireless coverage beyond terrestrial networks. Non-Terrestrial Networks (NTNs) combining Low Earth Orbit (LEO) satellites and Unmanned Aerial Vehicles (UAVs) offer promising solutions for remote and disaster-affected areas. However, efficient resource management in hybrid satellite-UAV systems requires accurate traffic prediction to handle dynamic topology changes, frequent handovers, and varying link conditions. We propose a long short-term memory (LSTM)-based deep learning framework that predicts network traffic using seven features, including satellite elevation, link type, SNR, latency, Doppler shift, and throughput. Our dual-layer LSTM architecture with batch normalization achieves 9.15 % MAPE, outperforming ARIMA by 36.5 % and vanilla RNN by 1.1 %. Experiments on real 5G traffic data demonstrate stable prediction up to 30 minutes ahead with 8.72 Mbps MAE, enabling proactive resource allocation in next-generation NTNs.

Fields of Study & Subjects

Satellite Communication Systems Software-Defined Networks and 5G UAV Applications and Optimization

Publication details

Journal / source
International Conference on Artificial Intelligence in Information and Communication (ICAIIC)
Pages
219–222
Publication date
24 February 2026
Language
EN