Deep Reinforcement Learning-Based Cooperative Rate Splitting for Satellite-to-Underground Communication Networks

Abstract

Reliable downlink communication in satellite-to-underground networks remains challenging due to severe signal attenuation caused by underground soil and refraction in the air-soil interface. To address this, we propose a novel cooperative rate-splitting (CRS)-aided transmission framework, where an aboveground relay decodes and forwards the common stream to underground devices (UDs). Based on this framework, we formulate a max-min fairness optimization problem that jointly optimizes power allocation, message splitting, and time slot scheduling to maximize the minimum achievable rate across UDs. To solve this high-dimensional non-convex problem under uncertain channels, we develop a deep reinforcement learning solution framework based on the proximal policy optimization (PPO) algorithm that integrates distribution-aware action modeling and a multi-branch actor network. Simulation results under a realistic underground pipeline monitoring scenario demonstrate that the proposed approach achieves average max-min rate gains exceeding $167\%$ over conventional benchmark strategies across various numbers of UDs and underground conditions.