Periodic Reporting for period 2 - SINAPSE (Software defined networking architecture augmented with Artificial Intelligence to improve aeronautical communications performance, security and efficiency)
Reporting period: 2021-05-01 to 2022-10-31
The SINAPSE proposes an intelligent and secured aeronautical datalink communications network architecture design based on the Software Defined Networking (SDN) architecture model augmented with Artificial Intelligence (AI) to predict and prevent safety services outages, to optimize available network resources and to implement cybersecurity functions protecting the network against digital attacks. SINAPSE has the following detailed objectives
• Objective 1: Design a solution suitable for ATM needs. This objective is mainly covered by WP1 activities. It starts by identifying relevant ATM operational, performance, safety and security requirements (D1.1 deliverable). It also consists of proposing a consolidated SDN augmented with AI design that complies with these requirements (D1.2)
• Objective 2: Guarantee ATC datalink services performance. This objective is covered in WP2 activities. It starts by designing and prototyping an AI application to anticipate and prevent service issues and outages (D2.1). It also consists of designing an SDN based aeronautical network integrating the ML (ML) application (D2.2)
• Objective 3: Optimize network resources. This objective is covered in WP3 activities. It starts by designing and prototype an AI application supporting QoS prediction to optimize network resources (D3.1). Additionally, it consists of designing an SDN based aeronautical network integrating the ML application (D3.2)
• Objective 4: Implement cybersecurity mechanisms to detect and prevent digital attack. This objective is covered by WP4 activities. It starts by designing and prototype an AI application for cybersecurity against prevalent network threat violating network confidentiality and integrity (D4.1). It also consists of Design a security architecture for SDN-based aeronautical network integrating the AI application (D4.2)
• Objective 1: Design a solution suitable for ATM needs. This objective is mainly covered by WP1 activities. It starts by identifying relevant ATM operational, performance, safety and security requirements (D1.1 deliverable). It also consists of proposing a consolidated SDN augmented with AI design that complies with these requirements (D1.2)
• Objective 2: Guarantee ATC datalink services performance. This objective is covered in WP2 activities. It starts by designing and prototyping an AI application to anticipate and prevent service issues and outages (D2.1). It also consists of designing an SDN based aeronautical network integrating the ML (ML) application (D2.2)
• Objective 3: Optimize network resources. This objective is covered in WP3 activities. It starts by designing and prototype an AI application supporting QoS prediction to optimize network resources (D3.1). Additionally, it consists of designing an SDN based aeronautical network integrating the ML application (D3.2)
• Objective 4: Implement cybersecurity mechanisms to detect and prevent digital attack. This objective is covered by WP4 activities. It starts by designing and prototype an AI application for cybersecurity against prevalent network threat violating network confidentiality and integrity (D4.1). It also consists of Design a security architecture for SDN-based aeronautical network integrating the AI application (D4.2)
The aeronautical communications systems are susceptible to temporary or permanent disruptions. These can be caused by local conditions, such as a high number of aircraft trying to send or receive data with ground operators, a lack of coverage between aircraft and ground centres or weather conditions degrading satellite communications (SATCOM). These systems are also vulnerable to cyberattacks, a growing concern within the aeronautical industry.
• Up until now, over the current ATN, it takes six minutes for a datalink disruption to be confirmed, a period during which controllers’ orders cannot be executed by pilots. SINAPSE assessed monitoring solution that uses artificial intelligence to predict such outages and that can be integrated into future aeronautical systems with prevention capabilities. Any disruption experienced by one aircraft has indirect implications for other aircraft in the vicinity, as airspace capacity is degraded, and planes may be delayed or re-routed. Any anticipation or prediction of these outages helps to reduce their impact on overall traffic. SINAPSE implemented a real-time operational data and network monitoring to predict communication failures using Controller Pilot Data Link Communications (CPDLC) data, captured in real-time from the operational ATN. A targeted use case demonstrated that SINAPSE could continuously predict and forecast disruption events ten minutes before they happened. This information could be very useful and could eventually prevent communication loss events in different ways.
• SINAPSE proposed an SDN design with a distributed software architecture that allows for increased configuration over the network, with everything monitored by a central controller layer. In traditional systems, this controller-like concept relies mainly on humans and is not suitable for automation, but SINAPSE introduced artificial intelligence as the controller, to manage the system more efficiently. The AI automatically checks for faults in the networks and using predictive information can proactively adjust the system and perform maintenance. SINAPSE consolidated the design of a Multi-layered hybrid hierarchical control plan structure that reduces the complexity growth of the SDN by partitioning the network into multiple segments and assigning several controllers to each to improve the scalability.
• An assessment of Machine Learning (ML) methodologies was also applied to predict the probability of transmission errors over satellite link communication that provides interactive voice and data telecommunication services for air traffic control and, as such, requires high availability and performance. Outages do not only result in a degradation of service, but they also constitute safety risks. The idea is to use these predictions to perform network optimizations. The ML model consisted of predicting the signal strength of a satellite link. The satellite link performance data was combined with associated regional weather data to create training dataset.
• SINAPSE Studied safety filter concept for data prediction responsible for deciding on the usability of the ML model predicted data. Safety filter works as a safeguard, without human intervention, and qualifies the predicted sensor data, as valid or invalid, by applying captured expertise rules. This concept contributes to making AI safer and keeps it in check.
• SINAPSE studied ML algorithms, in a federated machine learning architecture, to analyse the network traffic for signatures known to match cyberattacks. These models directly feed the IPDS as part of a defence-in-depth approach to protect the network against malicious traffic. The federated learning (FL) network architecture ensures that only the AI’s models are shared among users, without the need for underlying data to be shared—enhancing security further. This type of collaborative cybersecurity function will be a crucial building block for a secured future aeronautical communication infrastructure.
• Up until now, over the current ATN, it takes six minutes for a datalink disruption to be confirmed, a period during which controllers’ orders cannot be executed by pilots. SINAPSE assessed monitoring solution that uses artificial intelligence to predict such outages and that can be integrated into future aeronautical systems with prevention capabilities. Any disruption experienced by one aircraft has indirect implications for other aircraft in the vicinity, as airspace capacity is degraded, and planes may be delayed or re-routed. Any anticipation or prediction of these outages helps to reduce their impact on overall traffic. SINAPSE implemented a real-time operational data and network monitoring to predict communication failures using Controller Pilot Data Link Communications (CPDLC) data, captured in real-time from the operational ATN. A targeted use case demonstrated that SINAPSE could continuously predict and forecast disruption events ten minutes before they happened. This information could be very useful and could eventually prevent communication loss events in different ways.
• SINAPSE proposed an SDN design with a distributed software architecture that allows for increased configuration over the network, with everything monitored by a central controller layer. In traditional systems, this controller-like concept relies mainly on humans and is not suitable for automation, but SINAPSE introduced artificial intelligence as the controller, to manage the system more efficiently. The AI automatically checks for faults in the networks and using predictive information can proactively adjust the system and perform maintenance. SINAPSE consolidated the design of a Multi-layered hybrid hierarchical control plan structure that reduces the complexity growth of the SDN by partitioning the network into multiple segments and assigning several controllers to each to improve the scalability.
• An assessment of Machine Learning (ML) methodologies was also applied to predict the probability of transmission errors over satellite link communication that provides interactive voice and data telecommunication services for air traffic control and, as such, requires high availability and performance. Outages do not only result in a degradation of service, but they also constitute safety risks. The idea is to use these predictions to perform network optimizations. The ML model consisted of predicting the signal strength of a satellite link. The satellite link performance data was combined with associated regional weather data to create training dataset.
• SINAPSE Studied safety filter concept for data prediction responsible for deciding on the usability of the ML model predicted data. Safety filter works as a safeguard, without human intervention, and qualifies the predicted sensor data, as valid or invalid, by applying captured expertise rules. This concept contributes to making AI safer and keeps it in check.
• SINAPSE studied ML algorithms, in a federated machine learning architecture, to analyse the network traffic for signatures known to match cyberattacks. These models directly feed the IPDS as part of a defence-in-depth approach to protect the network against malicious traffic. The federated learning (FL) network architecture ensures that only the AI’s models are shared among users, without the need for underlying data to be shared—enhancing security further. This type of collaborative cybersecurity function will be a crucial building block for a secured future aeronautical communication infrastructure.
SINAPSE allowed the identification of SDN based network architectures, hybrid and hierarchical distributed SDN, suitable for Future ATN Communication infrastructure. The different AI methodologies (recursive, deep learning or regression etc.) and architectures (based on Federated Learning), suitable for cybersecurity and safety services outages prediction use cases, were also identified, and assessed. A datalink failure prediction solution was also validated by deploying a real-time datalink monitoring solution on Air Navigation Service Provider (ANSP) networks.
The next steps for SINAPSE will be to foster these concepts among ATN stakeholders in order emerge a collaborative initiative that will allow to share operational data, while respecting data privacy, to train AI models and to share and use these models among ATN communities. Additionally, there will be a need to develop and implement a demonstrator that enable an intelligent and fully automated security management targeted for the future ground ATN over IP network to mitigate cyber security attacks and initiate pilot projects in real production environments. Moreover, ML models certification and consistency should be also studied further, on how to keep a human in the loop and how to monitor AI predictions and keep it in check.
The next steps for SINAPSE will be to foster these concepts among ATN stakeholders in order emerge a collaborative initiative that will allow to share operational data, while respecting data privacy, to train AI models and to share and use these models among ATN communities. Additionally, there will be a need to develop and implement a demonstrator that enable an intelligent and fully automated security management targeted for the future ground ATN over IP network to mitigate cyber security attacks and initiate pilot projects in real production environments. Moreover, ML models certification and consistency should be also studied further, on how to keep a human in the loop and how to monitor AI predictions and keep it in check.