Contributions to aircraft trajectories anomaly detection and hyperbolic routing
Differentiating bird flight from UAS flight
Unmanned aircraft systems (UAS) are widely used for recreating, safety or video surveillance. They provide key information and reduce the risks associated with human operators. While UAS traffic nearly doubles yearly, the risks associated with UAS also increase. The concept of Urban Air Mobility, to autonomously fly people over cities, is currently addressed. Considering the exponential growth in UAS traffic, the risk of collision and privacy issues are also increasing. The Single European Sky Air traffic management Research Joint Undertaking (SESAR-JU) has developed the concept of U-Space, a set of services to enable the safe, secure and environmentally friendly use of drones. To support U-Space, an efficient method to detect and/or track UAS is essential for air traffic safety, and there exist several proposals attempting to overcome this challenge. Unfortunately, several of these schemes fail at differentiating birds from UAS.
The objective of this project aims to develop an algorithm capable to both detecting a UAS and identifying it against birds. The objective is to subsequently enable countermeasures against UAS. The proposed system will identify a UAS through a variety of methods, including Wavelet Linear Methods and Deep Learning. Once a UAS is detected, a countermeasure can be used with the tracking system. The final objective is to keep malicious or harmful UAS away from restricted or residential areas.
Currently available technologies will be reviewed for their suitability at fulfilling the performance requirements in terms of the ability to provide coverage at low levels, very low-level detection, and tracking together with the ability to correctly classify small drones from other targets (e.g. birds/manned aircraft/vehicles/people). The technology gaps will be identified, the ability to enhance key performance areas such as detection sensitivity of UAS in the presence of strong ground clutter), tracking and robust classification methods will be assessed, and an enhanced concept will be developed.
Detecting abnormal flight drift using deep learning
The radar traces carried within the ATM (Air Traffic Management) system allows civil aviation controllers to perform air traffic control and thus ensure the safety of aircraft and their passengers. However, the recent emergence of attacks from the Internet world with the high criticality of these data challenges the integrity of radar data. Furthermore, if we let aside the data, detecting a bypass drift of an aircraft flight from an abnormal/suspicious drift is also a challenging problem. Although state-of-the-art proposes some solutions to protect against flooding or potential man-in-the-middle attacks, the anomaly prediction schemes currently used in the world of IP networks cannot be transposed to avionics systems and radar traces. In addition, despite the criticality characterising radar exchanges, detecting anomalies related to these data has been little studied until now.
Using novel approaches in Deep Learning combined with Wavelet Linear Methods, this project aims to propose an algorithm able to predict the long-term behaviour of a flight radar trace and classify an abnormal route flight drift weighted by confidence intervals. The idea is to provide an algorithm that is generic enough to be applied to any context to analyse any suspicious vehicle drift on land, air or sea.
Routing over LEO megaconstellations using hyperbolic geometry
Satellite constellations have become an essential means of communication since private companies deploy several thousand satellites in low earth orbit (LEO), such as SpaceX with Starlink, Amazon with Kuiper or Eutelsat-OneWeb. The main objective of these massive constellations is to provide global terrestrial coverage. This increase in connectivity reduces the digital divide and allows new connectivity offers, such as those proposed by OneWeb for airliners.
The communication within a constellation is established in the following way. First, the signal from a ground station is sent to a satellite that sends the signal back to another ground station and so on, before reaching the final destination. Then, to establish communication between the ground stations, a path must be established on which the IP packets will be relayed. A routing protocol is then used within the constellation, specifying for each satellite the next hop to reach the destination satellite. However, routing in these constellations remains a complex and open problem. In recent years, many scientific publications have proposed new routing paradigms or improvement algorithms.
This project aims to explore a new method of path establishment based on the principle of hyperbolic geometry and, more particularly, by applying some properties of the Poincaré disk. Many distributed routing schemes, such as LEO routing, rely on greedy algorithms. Among these routing techniques, the simplest ones are based on geographic coordinates. This is called “greedy routing” because nodes always forward their messages to the neighbour closest to the destination in Euclidean distance.
However, this type of algorithm can go wrong when there is a node closer to the destination than all its neighbours without being the destination. This node is referred to as the minimum local, and packets passing through this node will not reach the given destination. To avoid local minima, another solution is to define an overlay or graph embedding. An overlay is a graph embedded in a network with its own notion of distance between elements. An overlay is said to be greedy if and only if the greedy routing is always valid and state-of-the-art, proving that any finite connected graph has a greedy overlay in the hyperbolic plane. The goal of this project will be to study the possibility of obtaining a greedy overlay that would allow a more efficient routing than the traditional short path based on the Floyd-Warshall algorithm commonly used for LEO routing.
Computer Sciences, Telecommunications, Artificial Intelligence, Networking, Applied Mathematics