LSTM-based Dynamic Routing with non-ISL LEO Satellite Constellations for Remote IoT Connectivity

This paper presents a novel sequence-to-sequence LSTM-based routing algorithm, designed for delay-sensitive remote IoT communication via both terrestrial and Low Earth Orbit (LEO) satellite relays when no inter-satellite links are available. We introduce a matrix-based approach to compute the communication delay between pairs of IoT nodes and satellites while considering their orbital passes and temporary visibility. Then, we develop an encoder-decoder architecture based on LSTM networks augmented with a beam-search strategy that enables the efficient prediction of optimal routing paths. Through experimental evaluations, incorporating a K - 2 beam search optimization, the algorithm demonstrates superior performance in generating near-optimal and scalable routes. Comparative analyses indicate that this approach outperforms traditional routing methods, offering lower data delays with reduced computational overhead.