Performance analysis of Q-learning-based NOMA in Satellite–Terrestrial Relay Networks

In this paper, we analyze the performance of $Q$-learning-based Non-Orthogonal Multiple Access (NOMA) in Satellite–Terrestrial Relay Networks (STRNs), addressing key challenges in massive Internet of Things (IoT) communications. Specifically, we focus on energy efficiency and normalized throughput metrics in uplink scenarios. By integrating a distributed $Q$-learning algorithm with NOMA, IoT devices can autonomously optimize transmission parameters – such as time slots, channels, and power levels – enhancing overall network performance. The proposed scheme outperforms fixed-power strategies by achieving higher normalized throughput and energy efficiency under varying network densities, offering up to 73% improvement in energy efficiency. Simulation results validate the protocol’s effectiveness, demonstrating its potential for large-scale IoT deployments in STRNs through efficient power allocation and reduced collision rates.