LEO Satellite Sensing Using Support Vector Machine Over Riemannian Manifolds

Today, both satellite and terrestrial networks are experiencing significant growth in order to address the rising demand for enhanced throughput, reduced latency, and seamless connectivity. Integrated satellite-terrestrial networks (ISTNs) have emerged as a promising solution to achieve global coverage, particularly in the context of sixth-generation (6G) mobile communication networks. However, the limited availability of spectrum presents a significant challenge to the persistent development of ISTNs. To address this issue, spectrum sensing plays a critical role in enabling satellite and terrestrial networks to dynamically share their spectrum resources. In this paper, we propose a support vector machine (SVM)-based learning model for sensing low-Earth-orbit (LEO) satellite signals over Riemannian manifolds. To classify the presence or absence of LEO satellites at a terrestrial base station, we utilize the sample covariance matrices of received signals, which reside on Riemannian manifolds due to their symmetric positive definite (SPD) properties. Simulation results validate the effectiveness of utilizing SVM model over Riemannian manifolds in terms of the detection probability and the probability of false alarm. Even under low signal-to-noise ratio regions, the system achieves full detection capability when only large-scale fading is considered. Additionally, when small-scale fading is included, the detection accuracy remains close to 90%. Furthermore, our proposed model exhibits remarkable efficiency, requiring only 40% of the total 3000 samples to train the model and achieve comparable performance with the standard approach of 80%–20% training-testing split.