A Predictive Approach for Transport Protocol Multi-Connectivity Scheduling

In this era of 5G and beyond, Non-Terrestrial Networks (NTN) plays a pivotal role in providing ubiquitous connectivity. Hence, it is essential to maintain connectivity with NTN, although a device is already connected to a Terrestrial Network (TN). The ideal approach would be to handle this at transport protocol layer as independent networks are involved. Establishing Multiple-Paths (MPs) between NTN and TN by means of MPTCP is a promising approach to realise this. Both the networks have different data rate and the available capacity, which varies over time based on the traffic demand. In this work, we start by presenting a solution to predict the traffic demand on NTN link based on federated Kolmogorov Arnold networks (fed-KANs). This predicted traffic demand is then used as an input to the MP scheduler to efficiently utilize the bandwidth available on both the NTN and TN. The goal is to minimize the data file transfer time and increase throughput. The proposed scheduling mechanism reduces the file transfer time compared to benchmark MP schedulers and achieves higher aggregated throughput. This shows that a scheduling mechanism with the knowledge of expected traffic demand performs better than classical schedulers which do take this into consideration.