CQI Feedback-Driven Regression for Interference-Aware SINR Prediction in LEO Mega-Constellation Networks

Low Earth Orbit (LEO) mega-constellation networks constitute a critical component in 6G architectures, enabling global coverage and low latency communications. However, their dense constellation deployment and overlapping beam patterns and aggressive spectrum reuse induce severe co-channel interference, significantly degrading downlink signal-to-interference-plus-noise ratio (SINR). This paper proposes a novel CQI (Channel Quality Indicator)-feedback-driven regression framework for joint interference identification and SINR prediction in LEO satellite downlinks. Our methodology establishes a nonlinear regression model that leverages historical network operation data to characterize interference relationships between interested user equipment and any of its interfering spot beams. To address the inherent granularity limitations of CQI feedback, we develop a semi-supervised pseudo-labeling mechanism integrated with transfer learning, which achieves enhanced interference pattern recognition while maintaining low air interface overhead.