By Eva Lagunas, Madyan Al-Senwi, Victor Monzon-Baeza and Jorge Querol, University of Luxembourg, Luxembourg
Research in Artificial Intelligence (AI) has been active for several decades, but lately, and with the exponential increase in the amounts of available data, new scenarios and use cases have emerged and have particularly influenced the field of wireless communications. AI has recently received significant attention as it is considered a fundamental technology for the next generation of 5G/6G cellular networks. In an attempt to achieve a real global connectivity and a fair access to internet over the globe, Non-Terrestrial Network (NTN) are expected to integrate the terrestrial infrastructure in a seamless and efficient manner. Furthermore, NTNs can offer back-up solution in case of unexpected connectivity failure, e.g. in natural disaster situation; or off-load traffic from congested terrestrial network.
Considering the above, AI presents as an excellent tool to manage both ground and space components of integrated TN-NTN networks. The entire system is expected to be massive in terms of network entities, consisting of satellites, ground stations, and end users. Such set-up makes it a very complex system where resources must be allocated optimally and dynamically. Typical non-AI solutions show some limitations is such scenarios:
- Extremely reduce decision-making time: If there is a word that characterize both TN and NTN is dynamism, i.e. the variability over time of both network topology as well as the time-variant traffic requests of the end-users. The latter combined with the typical non-convex or combinatorial nature of the resource management problem (e.g., carrier assignment, user scheduling) results in a critical challenge in terms of computational times for optimization-based solutions.
- Conventional communication system design relies on mathematical models, which are only accurate under strict assumptions. For instance, transmitter and receiver design typically are achieved under the assumption of accurate mathematical channel model knowledge. The imperfect channel estimation as well as the delays and impairments on the feedback channel can cause performance degradation to model-based design.
AI is investigated in TRANTOR as a promising technology with scalability and adaptability capabilities that can overcome such complex challenges. In particular, AI techniques can handle large volumes of data, adapt to new information, and improve performance over time, making them uniquely suited to the evolving landscape of integrated networks.