Probabilistic Link Opportunity Prediction for GNSS-Free NB-IoT Ground Devices in LEO Satellite Networks

Direct-to-satellite Narrowband Internet of Things (NB-IoT) over Low Earth Orbit (LEO) constellations enables wide-area connectivity for low-power devices in remote environments. However, many ground devices operate without Global Navigation Satellite System (GNSS) receivers and cannot propagate orbital state on board to predict satellite visibility autonomously. This letter proposes a probabilistic framework to predict link opportunities for GNSS-free NB-IoT devices using historical Two-Line Element (TLE) data. The problem is formulated as a binary classification over orbital geometry and time-derived features. Random Forest (RF) and Long Short-Term Memory (LSTM) predictors are evaluated against deterministic geometry-based baselines. Results show that data-driven models capture temporal patterns and provide tunable decision thresholds that complement deterministic visibility estimation, supporting energy-efficient transmission scheduling for GNSS-free IoT devices under intermittent LEO connectivity.