This paper proposes an energy-optimized uplink resource allocation framework for 6G massive Internet of Things (IoT) networks assisted by a Simultaneous Transmitting and Reflecting Reconfigurable Intelligent Surface (STAR-RIS). Unlike prior works that optimize radio resources and STAR-RIS coefficients separately, we jointly control transmit power, subchannel assignment, and the full set of STAR-RIS amplitude splitting and phase-shift coefficients using a single Soft Actor-Critic (SAC) agent with Gumbel-Softmax relaxation. The resulting policy is trained offline in a centralized manner and executed online with edge cloud coordination. Extensive simulations based on 3GPP Urban Micro channels with up to 200 devices and a 128-element STAR-RIS show that the proposed framework achieves 24.3% higher energy efficiency, 18.7% higher aggregate throughput, 19.1% lower latency, and 21.6% longer network lifetime compared to state-of-the-art successive convex approximation baselines, while maintaining near-optimal fairness. The results demonstrate that tight cross-layer integration of propagation control and radio resource allocation via deep reinforcement learning is a scalable and effective solution for green 6G massive machine-type communications.
Kamal et al. (Mon,) studied this question.