Resilient MANET: Resilience-Driven Topology Optimization for 3D UAV Networks

This repository contains the research prototype and simulation framework for resilience-driven adaptive topology configuration in 3D UAV-based mobile ad hoc networks (MANETs / FANETs).
The work focuses on network reliability and resilience, rather than solely throughput or UAV-centric optimization.

This work has been accepted by ACM Transactions on Internet Technology (TOIT), Vol. 26, No. 1, 2025.
📄 Paper link: https://dl.acm.org/doi/10.1145/3747350

---

Overview

Unmanned aerial vehicles (UAVs) are increasingly used as dynamic networking components to extend coverage and connectivity in challenging environments. However, existing approaches often emphasize UAV-specific optimization or throughput maximization, while overlooking network-level reliability and resilience, especially under dynamic conditions such as user mobility, signal blockage, and topology disruptions.

This project proposes a topology-driven, resilience-aware framework for 3D UAV networks, leveraging scene-based information and adaptive topological (re)configuration to ensure reliable communication performance.
Rather than treating UAV positioning, connectivity, and evaluation independently, we adopt a holistic system view of network resilience.

---

Key Idea

We decompose the reliability challenge in 3D UAV networks into three tightly coupled stages:

1. Topological Resilience Quantification
   Quantifying network reliability using functional metrics, including:

   • Data rate and bandwidth delivery
   • Backup path availability
   • Network partitioning risk
   • Load distribution and overload conditions

2. UAV Self-Positioning (Position Finding)
   Determining UAV locations using scene-aware methods to improve coverage, connectivity, and resilience.

   • Heuristic and score-based approaches
   • Scene-based evaluation with dynamic ground users (GUs)

3. Learning-Based Connectivity Optimization
   Adapting GU–UAV–BS connectivity to improve resilience and load balancing under dynamic conditions.

Although learning-based methods were explored, this project emphasizes functional resilience and system-level behavior, rather than treating deep reinforcement learning (DRL) as a universal solution.

---

Methodology (from the Paper)

Our framework evaluates network resilience from a functional perspective, focusing on the system’s ability to consistently deliver high-quality communication while mitigating disruptions.

Key components include:

• Resilience Metrics

  • Resilience Score (RS)
  • Data Rate (DR)
  • Backup Path (BP)
  • Network Partition (NP)
  • Overload (OL)

• Scene-Based Simulation

  • Obstacle-rich 3D environments
  • Dynamic GU mobility
  • UAV-to-UAV and UAV-to-BS interactions

• Adaptive Reconfiguration

  • UAV repositioning
  • Connectivity adjustment
  • Load-aware topology evolution

Extensive experiments demonstrate significant improvements over baseline methods in GU capacity, UAV load balancing, and adaptability under dynamic conditions.

---

Resilience Score (RS) Quantification

To evaluate network reliability from a functional and topology-aware perspective, we define a Resilience Score (RS) that jointly captures bandwidth efficiency, backup path robustness, and network partition tolerance.

Overall Resilience Score

The overall resilience score is defined as a weighted combination of three components:

$$ RS = w_1 \cdot R_b + w_2 \cdot R_{bp} + w_3 \cdot R_{np}, $$

where:

• $R_b$ denotes the bandwidth allocation score,
• $R_{bp}$ denotes the backup path reservation score,
• $R_{np}$ denotes the network partitioning score,
• $w_1, w_2, w_3$ are weighting factors reflecting the relative importance of each component.

This formulation enables a holistic evaluation of network resilience under dynamic topology and node failures.

---

Bandwidth Allocation ($R_b$)

The bandwidth allocation score evaluates communication efficiency and fairness from base stations (BSs) to ground users (GUs) via UAVs. It integrates both worst-case and average user throughput:

$$ R_b = r \cdot \min_{i \in U} \left( 1, \frac{R_{i,\min}}{\Delta F} \right)+ (1-r) \cdot \min_{i \in U} \left( 1, \frac{R_{i,\text{avg}}}{\Delta F} \right), $$

where:

• $R_{i,\min}$ and $R_{i,\text{avg}}$ are the minimum and average data rates across GUs,
• $\Delta F$ is the achievable bandwidth of a UAV,
• $r \in [0,1]$ is a penalty parameter controlling the trade-off between fairness and efficiency.

The individual user rate $R_i$ is computed using the Shannon capacity formula, where the received power at GU $i$ from UAV $u$ is given by:

$$ P^{Rx}_{u,i} = P^{Tx}_u \cdot G_u \cdot G_i \cdot \left( \frac{\lambda}{4 \pi d_{u,i}} \right)^n. $$

Here:

• $P^{Tx}_u$ is the UAV transmit power,
• $G_u$ and $G_i$ are antenna gains,
• $d_{u,i}$ is the 3D distance between UAV $u$ and GU $i$,
• $\lambda = \frac{c}{f}$ is the signal wavelength,
• $n$ is the path loss exponent ($n=2$ for LoS, $n=4$ for NLoS).

---

Backup Path Reservation ($R_{bp}$)

The backup path score quantifies the network’s ability to maintain connectivity when primary links fail:

$$ R_{bp} = \sum_{i \in GU} \sum_{p \in P_f} PS(i,p), $$

where $P_f$ is the set of filtered candidate backup paths.

The path score $PS(i,p)$ is defined as:

$$ PS(i,p) = \begin{cases} 1, & \text{if } B(p) \ge B(OP), \\ \frac{B(p)}{B(OP)}, & \text{if } B(p) < B(OP) \ &\ \Delta h \le 0, \\ \frac{B(p)}{B(OP)\cdot \Delta h}, & \text{if } B(p) < B(OP) \ &\ \Delta h > 0, \\ -1, & \text{otherwise}. \end{cases} $$

Here:

• $B(p)$ is the bottleneck link rate of path $p$,
• $B(OP)$ is the bottleneck rate of the optimal primary path,
• $\Delta h = h(p) - h(OP)$ is the hop difference between backup and primary paths.

This formulation favors backup paths with higher bandwidth and fewer hops, while penalizing unreliable alternatives.

---

Network Partitioning Score ($R_{np}$)

The network partitioning score evaluates resilience under UAV node failures:

$$ R_{np} = \alpha \cdot \frac{1}{|D|} \sum_{d_f \in D} \frac{R_b(d_f)}{R_b} + \beta \cdot \frac{1}{|D|} \sum_{d_f \in D} \frac{R_{bp}(d_f)}{R_{bp}}, $$

where:

• $D$ is the set of all possible UAV failure combinations,
• $d_f$ represents a specific failure scenario,
• $R_b(d_f)$ and $R_{bp}(d_f)$ are the degraded scores under failure,
• $\alpha + \beta = 1$.

By averaging performance degradation across failure cases, $R_{np}$ captures the network’s tolerance to node loss and partitioning.

---

Learning-Based Exploration (Q-Learning / DRL)

Learning-based methods were explored to validate whether topology optimization could be learned efficiently:

• Q-learning demonstrated that near-optimal topologies can be identified at very early stages, requiring limited computational resources.
• DRL-based UAV positioning was investigated but showed unstable or inferior performance compared to structured, scene-aware approaches.

As a result, learning is treated as a supporting or exploratory component, rather than the core dependency of the system.

---

Summary

Together, $(R_b, R_{bp}, R_{np})$ enable a functionally grounded, topology-aware resilience evaluation, supporting adaptive UAV positioning and connectivity optimization in dynamic 3D network environments.

Project Structure

This repository reflects an iterative research implementation and is organized as follows:

resilientMANET/
├── classes/                 # Core abstractions (Nodes, UAVMap, BackhaulPaths)
├── config/                  # Scene and system configuration (JSON)
├── functions/               # Utility functions and resilience quantification
├── csv_tables/              # Experimental result tables
├── functions/               # Utility functions and resilience quantification
├── paper_revision/          # Files used during paper revision
├── paper_revision_2/        # Files used during later revision stages
├── position_finding_DRL/    # Legacy DRL attempts (kept for reference)
├── q_tables/                # Q-tables generated during evaluation
├── baseline_simu.py         # Baseline evaluation entry point
├── system_simu*.py          # System simulation and variants
├── visualization_functions.py
└── README.md

Notes on Important Folders

classes/

Contains the three most important classes in this project:

• Nodes
  Defines the fundamental network entities (GU, UAV, BS) and their attributes.

• UAVMap
  Maintains the spatial and topological relationships among UAVs and ground users.

• BackhaulPaths
  Handles backhaul connectivity and path-related operations used in resilience evaluation.

---

functions/

Includes key utility functions, especially those for resilience quantification under dynamic ground user (GU) populations, such as bandwidth evaluation, backup path scoring, and network partition analysis.

---

config/

Contains scene configuration files in JSON format, defining:

• Environment layout and obstacles
• Ground users (GUs)
• UAVs and base stations (BSs)
• Simulation and evaluation parameters

---

position_finding_DRL/

Contains early DRL-based UAV position finding experiments.

These learning-based approaches were explored during the early stage of this project but did not outperform the main structured, scene-aware methods.
They are not part of the final system and are kept only for future reference.

---

paper_revision/ and paper_revision_2/

Temporary folders used during different stages of paper revision, including intermediate figures, tables, and experimental results.

---

Notes on Reproducibility

This repository reflects a research prototype developed during early-stage exploration.
The focus is on system behavior, resilience modeling, and evaluation methodology, rather than turnkey reproducibility.

• Baseline evaluation is performed via baseline_simu.py
• Scene configurations are defined in config/
• Experimental outputs are recorded as CSV tables and figures

---

Citation

If you use this work in your research, please cite:

@article{resilientMANET2025,
  title={Resilience-Driven Adaptive Topology Configuration for 3D UAV Networks},
  author={Jiayuan Huang, Mingzhe Chen, Yuchen Liu},
  journal={ACM Transactions on Internet Technology},
  volume={26},
  number={1},
  year={2025},
  doi={10.1145/3747350}
}