AAU Student Projects - visit Aalborg University's student projects portal
An executive master's programme thesis from Aalborg University
Book cover


Dynamic Resource Allocation in Space-Air-Ground Integrated Networks Using Deep Reinforcement Learning

Author

Term

4. term

Publication year

2025

Submitted on

Pages

55

Abstract

Dette speciale undersøger, hvordan ressourcer i rum–luft–jord integrerede netværk (SAGIN) kan fordeles dynamisk for at understøtte 6G‑visionen om global, pålidelig og lavlatens kommunikation. Udgangspunktet er, at stigende trafik og dækning i områder uden infrastruktur kræver intelligent radioressourcestyring og brugerassociering på tværs af satellit-, UAV- og terrestriske links. Specialets mål er at maksimere throughput og spektrumeffektivitet, overholde backhaul‑begrænsninger og samtidig reducere energiforbruget. Til dette formål udvikles en dyb forstærkningslæringsløsning baseret på Deep Q‑Network (DQN), som lærer beslutninger om ressourceallokering i et dynamisk miljø. Scenariet omfatter to LEO‑satellitter og tre UAV’er, der fungerer som basestationer og leverer forbindelse til brugere på jorden i områder uden kommunikationsinfrastruktur. Arbejdet bidrager med problemformulering, scenariomodellering og et DQN‑baseret agent‑ og miljødesign samt en simuleringsramme, der kan evaluere ydelsen. Af praktiske grunde afgrænses metoden til DQN. Den udleverede tekst indeholder ikke kvantitative resultater; disse behandles senere i rapporten.

This thesis investigates how to dynamically allocate resources in Space–Air–Ground Integrated Networks (SAGIN) to support the 6G vision of global, reliable, low‑latency connectivity. As traffic grows and underserved areas remain without infrastructure, intelligent radio resource management and user association across satellite, UAV, and terrestrial links become crucial. The objective is to maximize throughput and spectrum efficiency, satisfy backhaul constraints, and reduce energy consumption. To this end, the work develops a deep reinforcement learning approach based on a Deep Q‑Network (DQN) that learns resource allocation decisions in a dynamic environment. The study models a scenario with two LEO satellites and three UAVs acting as base stations to serve ground users in areas lacking communication infrastructure. Contributions include problem formulation, scenario modeling, and the design and implementation of a DQN agent and simulation environment to evaluate performance. Due to time constraints, the scope focuses on DQN. The provided excerpt does not report quantitative results; these are addressed later in the thesis.

[This abstract was generated with the help of AI]