A Novel AI-Driven Graph-Swarm THz Slice Optimizer for Terahertz Frequency Management and Network Slicing in 6G/7G ORAN Networks

Abstract

As 6G and 7G networks evolve, the efficient management of the Terahertz (THz) frequency band and network slicing in Open Radio Access Network (ORAN) architectures is critical to support ultra-high-speed data transmission, diverse service requirements, and dynamic network conditions. This research addresses key challenges such as interference management in the high-density THz spectrum, unpredictable traffic patterns, and fluctuating service demands in network slicing. The proposed AI-Driven Graph-Swarm THz Slice Optimizer Framework introduces two novel components: Co-Annealed Graph-AE OptiNet for real-time THz frequency optimization and Deep Q-Cat Memory Net for adaptive network slicing. The Co-Annealed Graph-AE OptiNet dynamically models the ORAN network state, predicts interference, and optimizes spectrum allocation, achieving 95.7% spectrum efficiency and 2.2% interference rates, ensuring minimal signal degradation. Simultaneously, the Deep Q-Cat Memory Net learns optimal slicing strategies, predicts congestion, and proactively allocates resources, resulting in 97.8 Gbps throughput, 0.8 ms latency, and improved bandwidth utilization. Simulation results validate the framework's effectiveness, showcasing significant improvements over existing models in all key performance metrics, including low latency, enhanced resource utilization, and robust adaptability. These findings highlight the framework's potential to enable scalable and efficient network management in future-generation wireless networks.