AI-DRIVEN FEDERATED MULTI-AGENT ACTOR–CRITIC LEARNING FOR SECURE AND ENERGY-EFFICIENT RESOURCE OPTIMIZATION IN 6G SEMI-GRANT-FREE NOMA-BASED IOT NETWORKS

Abstract

Sixth-generation (6G) wireless systems are expected to support ultra-massive machine-type communication through dense deployments of Internet of Things (IoT) devices. Semi-Grant-Free Non-Orthogonal Multiple Access (SGF-NOMA) has emerged as a promising access mechanism for enabling simultaneous transmissions of grant-based and grant-free users while improving spectral efficiency. Dense IoT environments create significant challenges including power-domain collisions, energy inefficiency due to repeated retransmissions, and susceptibility to malicious interference such as jamming attacks. Conventional centralized optimization techniques introduce high signaling overhead and raise privacy concerns, limiting their scalability in large-scale IoT deployments. This work proposes an AI-driven federated hybrid multi-agent actor–critic learning framework for secure and energy-aware resource optimization in 6G SGF-NOMA IoT networks. Each grant-free device operates as an intelligent learning agent that autonomously selects transmission power levels and resource blocks based on local observations. A federated learning architecture enables decentralized model training while periodically aggregating parameters at an edge server using federated averaging. The proposed framework integrates a hybrid exploration–cooperation strategy and a multi-objective reward function that jointly considers throughput gain, collision penalties, energy consumption cost, and security robustness under jamming interference. Simulation-based evaluations demonstrate that the proposed framework significantly improves conditional throughput, reduces power collision probability, and enhances energy efficiency compared with centralized deep reinforcement learning, random access, and conventional SGF-NOMA approaches. Results also indicate faster convergence and improved fairness among IoT devices under ultra-dense deployment scenarios. The proposed solution provides a scalable and privacy-preserving learning architecture suitable for AI-native 6G wireless systems and large-scale IoT networks.