Intelligent Multipath Routing for Multimedia Services over Unmanned Aerial Mobility Networks = 무인 항공 이동 네트워크 환경에서 멀티미디어 서비스를 위한 지능형 다중경로 라우팅 기법

다국어 초록 (Multilingual Abstract)

Drone communications are increasingly becoming a crucial complement to existing Internet infrastructure, enabling user services with seamless connectivity that was previously infeasible. However, engineering network traffic dynamically across multiple drone-to-drone links poses significant challenges for both energy efficiency and reliable performance, and intelligent video streaming in dense drone swarms is especially difficult given probabilistic access to cellular infrastructure and a highly dynamic network topology. To address these described challenges, we propose a software-defined networking （SDN） approach. Uisng centralized routing to control drone swarms, inte- grating geographic routing, queue-aware water flow scheduling, and deep reinforcement learning （DRL） for adaptive path optimization. A policy trained using Proximal Policy Optimization （PPO） predicts routing decisions from network state observations including queue depths, energy levels, and connectivity, jointly optimizing routing paths and link scheduling for scalable, energy-aware, and delay-sensitive streaming. Simulation results using ns-3 demonstrate substantial improvements in throughput and packet delivery ratio over traditional routing protocols （AODV, OLSR, DSDV）, with the SDN- RL approach achieving significant delivery-ratio gains compared to traditional protocols.

국문 초록 (Abstract)

드론 통신은 기존 인터넷 인프라의 중요한 보완 수단으로 자리잡고 있으며, 이전에는실현하기 어려웠던 원활한 연결성을 통해 다양한 사용자 서비스를 가능하게 하고있습니다. 그러나 다수의 드론 간 링크를 통해 네트워크 트래픽을 동적으로 제어하는것은 에너지 효율성과 안정적인 성능을 동시에 달성하는 데 상당한 어려움을야기하며, 특히 밀집된 드론 군집에서의 지능형 비디오 스트리밍은 셀룰러 인프라에대한 확률적 접근과 매우 동적인 네트워크 토폴로지로 인해 더욱 어렵습니다.이러한 문제를 해결하기 위해 본 논문은 드론 군집을 위한 중앙 집중식 라우팅제어 기반의 소프트웨어 정의 네트워킹（SDN） 접근 방식을 제안하며, 지리적 라우팅,큐 인식 워터플로우 스케줄링, 그리고 적응형 경로 최적화를 위한 심층 강화학습（DRL）을 통합합니다. 근접 정책 최적화（PPO）로 학습된 정책은 큐 길이, 에너지수준, 연결성을 포함한 네트워크 상태 관측으로부터 라우팅 결정을 예측하며, 확장가능하고 에너지를 고려하며 지연에 민감한 스트리밍을 위해 라우팅 경로와 링크스케줄링을 공동으로 최적화합니다.ns‑3를 사용한 시뮬레이션 결과는 기존 라우팅 프로토콜（AODV, OLSR, DSDV）대비 처리량과 패킷 전달률에서 상당한 개선을 보여주며, SDN RL 접근 방식은기존 프로토콜 대비 패킷 전달률에서 큰 개선을 달성합니다.

목차 (Table of Contents)

• 1 Introduction 1
• 1.1 Motivation 3
• 1.2 Contributions 4
• 2 Related Work 6
• 2.1 Energy-Aware UAV Operation and Deployment 6
• 2.2 Multipath Routing and Network Resilience in UAV Swarms 7
• 2.3 Traditional Routing Protocols for Ad-Hoc Networks 9
• 2.4 Software-Defined Networking for Aerial Networks 10
• 2.5 Position of This Work 11
• 3 Problem Statement 13
• 3.1 Channel Model and Link Capacity 14
• 3.2 Flow Conservation, Latency, and Energy Consumption 17
• 3.3 Queueing Dynamics and Mobility Model 19
• 3.4 Problem Formulation 21
• 4 Intelligent SDN-Based Routing Framework 23
• 4.1 SDN Controller Architecture 23
• 4.1.1 Design Principles 23
• 4.1.2 Controller Components 24
• 4.1.3 Relationship Among the Three Routing Protocols 25
• 4.2 Geographic Routing Protocol 26
• 4.2.1 Algorithm Description 27
• 4.2.2 Properties and Limitations 27
• 4.3 Backpressure based routing protocol 28
• 4.3.1 Backpressure Computation 29
• 4.3.2 Combined Routing Metric 29
• 4.3.3 Per-Epoch Forwarding Decisions 30
• 4.3.4 Properties 30
• 4.4 Reinforcement Learning-based Routing Protocol 31
• 4.4.1 Motivation for a Hybrid Design 32
• 4.4.2 State Space 32
• 4.4.3 Action Space 33
• 4.4.4 Reward Function 34
• 4.4.5 Policy Architecture and Training 35
• 4.4.6 Algorithm Summary 37
• 4.5 Feasibility Bound and Projection 37
• 4.6 Complexity Analysis 39
• 5 Performance Evaluation 40
• 5.1 Simulation Environment 40
• 5.1.1 Network Topology 40
• 5.1.2 Simulation User Interface 42
• 5.2 Compared Algorithms 42
• 5.2.1 Traditional Protocols 43
• 5.2.2 Proposed SDN Protocols 43
• 5.3 Experimental Results 44
• 5.3.1 Overall Performance Comparison 44
• 5.3.2 Time Series Throughput Analysis 47
• 5.3.3 Packet Delivery Ratio Analysis 48
• 5.3.4 End to End Delay Analysis 49
• 5.3.5 Scalability Analysis 50
• 5.3.6 Operating-Regime Dependence 51
• 5.3.7 Stream Stability （Variance Analysis） 53
• 5.3.8 Energy Consumption 55
• 5.3.9 Multidimensional Performance Comparison 55
• 5.3.10 Network Connectivity Analysis 55
• 6 Conclusion 57