TL;DR
This paper introduces a novel deep learning-based routing approach for aeronautical ad-hoc networks (AANETs) that leverages real flight data. It tackles the challenge of highly dynamic network topologies by enabling aircraft to make routing decisions using only local information, achieving performance close to optimal global routing and effectively navigating multi-objective optimization goals like delay, capacity, and lifetime.
What Problem Does This Paper Address?
Aeronautical ad-hoc networks (AANETs) are envisioned to provide ubiquitous internet access by using commercial aircraft as relays. However, the extremely high speeds and constantly changing positions of these aircraft create a highly dynamic network topology, making traditional routing protocols inefficient. Existing position-based routing methods struggle with "communication voids" (where no neighbor is closer to the destination) and often optimize only a single metric (like delay). Furthermore, multi-objective optimization (MOO) for routing, which considers multiple performance metrics simultaneously (e.g., delay, capacity, lifetime), typically requires global network knowledge, which is impractical in AANETs. This paper aims to bridge this gap by developing a deep learning-based routing solution that operates efficiently with local information and can handle both single-objective and multi-objective routing scenarios.
Methods
The core of the proposed solution is a Deep Neural Network (DNN). For single-objective routing (minimizing end-to-end delay), a DNN is trained offline using historical flight data. This DNN takes local geographic information (positions of the current node, its neighbors, and the destination) as input and outputs the estimated minimum delay from each neighbor to the destination. This allows the forwarding node to select the optimal next hop based solely on local data.
The approach is extended to multi-objective routing (minimizing delay, maximizing capacity, and maximizing lifetime). Here, a more complex DNN is employed, taking into account not only geographic information but also the desired capacity and lifetime constraints. The training process for both single and multi-objective scenarios utilizes real flight data to capture the regular mobility patterns of aircraft. A feedback mechanism is also introduced to improve the accuracy of delay estimations in the multi-objective case by recursively exploiting the optimal substructure.
Datasets
The paper utilizes real flight data collected over the North Atlantic Ocean and the European continent. This data includes timestamps, latitude, longitude, altitude, speed, and heading for flights. To augment the limited real data, synthetic flight data is generated by randomly shifting flight timelines. This synthetic data is then split into training, validation, and testing sets.
Key Insights and Ideas
- Leveraging Regular Mobility Patterns: The inherent predictability of commercial flight paths is a key enabler. By training DNNs on historical data, the network can learn to anticipate and adapt to these patterns, even with local information.
- Local Information Sufficiency: The research demonstrates that with a well-trained DNN, routing decisions can be made effectively using only local geographic information, eliminating the need for complex global state maintenance.
- Bridging the Gap to Pareto Optimality: The DL-based multi-objective routing approach can approximate the Pareto front (the set of optimal trade-offs between objectives) that would typically be found using global link information.
- Handling Communication Voids: The DNN's ability to learn complex relationships allows it to infer information that helps bypass or navigate communication voids more effectively than traditional greedy methods.
Key Results
- Single-Objective Routing: The proposed DL-aided routing significantly outperforms existing position-based routing protocols (like GPSR and GLSR) in terms of End-to-End (E2E) delay, achieving performance close to optimal routing that relies on global link information.
- Multi-Objective Routing: The DL-aided multi-objective routing algorithm successfully finds paths that approach the Pareto front, demonstrating effective trade-offs between delay, capacity, and lifetime. It also guarantees the specified lifetime constraints, which traditional methods like GLSR and GPSR struggle with.
- Computational Efficiency: While the multi-objective DNN has a higher training complexity, its one-shot evaluation time for online decision-making is very low (under 1 ms), making it practical for real-time applications.
Limitations and Future Work
- Flight Density Dependence: The paper acknowledges that AANETs may struggle in regions with low flight density, where forming a robust network might be challenging.
- MOO Complexity: Multi-objective optimization inherently increases computational complexity and processing time.
- Future Work: The authors suggest integrating satellites into AANETs to support truly global coverage. They also aim to further reduce the complexity of MOO for more efficient routing.
Summary updated