Skip to content
Home » News » Project » Technologies » Artificial Intelligence for Multi-Connectivity in Beyond-5G Non-Terrestrial Networks

Artificial Intelligence for Multi-Connectivity in Beyond-5G Non-Terrestrial Networks

By Pietro Cassarà, Alberto Gotta, and Achilles Machumilane, CNR-ISTI, Italy

In recent decades, the volume of Internet traffic has increased enormously, overwhelming the existing communications infrastructure. At the same time, technological development has led to new applications that have different and stringent requirements in terms of data rates, latency, reliability, and bandwidth that cannot be met by current transmission technologies. This challenge is expected to intensify with the ongoing global deployment of the fifth generation (5G) and the upcoming sixth generation (6G) mobile networks, which have more demanding applications and use cases such as Augmented Reality (AR), Extreme Reality (XR), fully autonomous driving, Internet of Things (IoT), Internet of Everything (EoE), Industrial IoT (IIoT) and holographic communication.

Such applications will require extremely high reliability, low latency, and very high data rates, which necessitate new communication technologies, architectures, and standards. Multi-connectivity (MC) has been proposed as one of the promising solutions for the stringent application requirements envisaged beyond 5G/6G mobile networks. MC is a communication technology that allows for concurrent connection over multiple paths of the same network or different networks to increase transmission efficiency by taking advantage of the different channel characteristics of each path. Transmission over multiple paths can improve channel availability and reliability, throughput, and data rates. It also allows for bandwidth aggregation, load balancing, and fault tolerance because using multiple paths for the same data stream makes short and intermittent path failures transparent to end users. The major challenge in MC is multipath scheduling. Multipath scheduling (MS) in MC networks is a process of selecting appropriate paths and determining the required amount of traffic to be assigned to each path. This requires knowledge of real-time Channel State Information (CSI), which may be difficult to obtain.

In wireless Non-Terrestrial Networks, MS is more challenging due to the temporal and spatial variability and heterogeneity of wireless channels and the mobility of network nodes. Moreover, when MC is coupled with Packet Duplication (PD) for traffic protection, estimating the required redundancy exacerbates the MC challenge because too little redundancy cannot provide the required protection, while too much redundancy wastes bandwidth. Thus, there is a trade-off between protection and bandwidth utilization. Unfortunately, most conventional schedulers and traffic protection techniques either emphasize one aspect and neglect the other or use static scheduling policies that do not consider the heterogeneity and dynamics of wireless networks, while others add computational overhead or retransmission delays, which degrade the quality of real-time multimedia traffic.

The goal of the TRANTOR project is to address these technology gaps by using Artificial Intelligence (AI) to support MC beyond 5G/6G networks. We aim to create an intelligent scheduling system that can exploit real-time path conditions to enhance traffic scheduling and meet predefined Quality of Service (QoS). To this end, we are using Reinforcement Learning (RL) and Generative Models (GMs) to develop an MS system that can learn in real-time the channel conditions of the available network paths and select an appropriate subset of the paths to overcome traffic losses by increasing the probability of correct traffic delivery. To protect traffic from path impairments and increase throughput, we couple MC with PD and route information traffic and corresponding duplicates over different paths so that information lost on one path can be recovered on the other paths. Therefore, such a scheduling system performs three main scheduling tasks: path selection, traffic allocation, and redundancy estimation.

Since it is a learning-based system, its scheduling decisions are based on real-time path information, making it able to adapt to varying path conditions dynamically. Our system can improve network reliability, service availability, data rates, and the overall user Quality of Experience (QoE). We used an Actor-Critic (AC)-RL algorithm to predict the LOS of multiple satellites visible to the User Equipment (UE) and schedule traffic on appropriate links to overcome traffic loss due to LOS fluctuations. However, it was found that in an MC system with many satellite links, the RL agent may take a very long time to converge compared to the satellite visibility time, which is usually very short, in the order of a few minutes. To address this problem, we extended AC-RL with Generative Models for efficient LOS estimation and multipath traffic scheduling, which uses generative AI to speed up the learning process of LOS estimation. This allows the agent to have a complete view of all the states of the available links at each learning event. This speeds up the learning process and allows the RL agent to converge within the satellite visibility period by increasing the convergence rate by 47%. Since it uses RL, this system outperforms selected legacy schedulers such as Round Robin (RR) and Weighted Round Robin(WRR), which use static scheduling policies that cannot cope with network fluctuations.

The use of adaptive PD makes our traffic protection strategy better for constrained devices compared to state-of-the-art traffic protection schemes. Since it can also predict LOS, our system can support MC in both TN and TN-NTN integrated networks, which are expected to shape the communication infrastructure beyond 5G/6G networks. In fact, reducing network analytics is mandatory to increase the scalability of an MC system with multiple links and reduce the amount of data that an agent has to collect, elaborate, and store.

This is TRANTOR, the foundation of a new set of technologies for 6G-NTNs.