A NOVEL MACHINE LEARNING TECHNIQUE FOR PREAMBLE DETECTION IN SATELLITE RANDOM ACCESS CHANNELS

Preamble detection is a critical task in satellite random access (RA) systems such as Enhanced Spread Spectrum Aloha (E-SSA). Conventional threshold-based techniques often struggle in dynamic environments like Low Earth Orbit (LEO) or under high terminal loads, where variations in power, delay, and frequency offset degrade cross-correlation outputs and lead to high false alarm (FA) and missed detection (MD) rates. Adaptive thresholding can improve detection but requires computationally expensive iterative optimization, making it unsuitable for resource-constrained platforms. To address this, we propose an efficient machine learning (ML)-based approach that reformulates preamble detection as a binary classification problem. By extracting relevant features from cross-correlation outputs and using a decision tree classifier (DTC), the method achieves higher robustness and lower complexity than conventional techniques. Simulation results based on VDE-SAT Link ID 20 waveform demonstrate significant gains in FA and MD performance, supporting the feasibility of ML-based detection in modern satellite networks.