Study of Cluster-Based Routing Based on Machine Learning for UAV Networks in 6G

TL;DR

This work tackles dynamic clustering and CH selection in FANETs for 6G by marrying mobility-prediction with ML-guided optimization. It uses XGBoost to forecast UAV positions and a composite CH-selection objective that balances intra-cluster distance and link power via a weight $w \in [0,1]$, operationalized through a practical scoring heuristic. The framework is evaluated in both centralized (5G) and distributed (6G) topologies with realistic video traffic, showing substantial gains in delay, jitter, and throughput in the distributed setting, while also highlighting potential jitter challenges in centralized deployments. Overall, the results validate ML-assisted clustering as a viable approach to scalable, reliable UAV networking in future 6G environments, with avenues for extension to GNNs, federated learning, and multiobjective optimization.

Abstract

The sixth generation (6G) wireless networks are envisioned to deliver ultra-low latency, massive connectivity, and high data rates, enabling advanced applications such as autonomous {unmaned aerial vehicles (UAV)} swarms and aerial edge computing. However, realizing this vision in Flying Ad Hoc Networks (FANETs) requires intelligent and adaptive clustering mechanisms to ensure efficient routing and resource utilization. This paper proposes a novel machine learning-driven framework for dynamic cluster formation and cluster head selection in 6G-enabled FANETs. The system leverages mobility prediction using {Extreme Gradient Boosting (XGBoost)} and a composite optimization strategy based on signal strength and spatial proximity to identify optimal cluster heads. To evaluate the proposed method, comprehensive simulations were conducted in both centralized (5G) and decentralized (6G) topologies using realistic video traffic patterns. Results show that the proposed model achieves significant improvements in delay, jitter, and throughput in decentralized scenarios. These findings demonstrate the potential of combining machine learning with clustering techniques to enhance scalability, stability, and performance in next-generation aerial networks.

Figures (8)

• Figure 1: System workflow for network clustering and routing simulation.
• Figure 2: Network topology designs.
• Figure 3: Objective function $J_i(w)$ for each candidate head as a function of the weight $w$.
• Figure 4: Complexity of cluster-head selection per cluster as $M$ grows (log-scale). The proposed pairwise signal-strength scoring scales quadratically; a fast spatial approximation using k-d trees + $k$NN trends toward $O(M\log M + kM)$. Metaheuristics are linear in $M$ but with large constants from iterations/populations, while MILP/exhaustive search grows super-exponentially.
• Figure 5: Average delay for all stations.