MAPPO-assisted Joint Task Offloading and Resource Allocation for Energy Efficiency in Dynamic NOMA-MEC Systems

Mobile Edge Computing (MEC) improves how well devices work and saves energy by moving computing tasks to nearby servers, while Non-Orthogonal Multiple Access (NOMA) makes better use of available bandwidth, making NOMA-MEC networks a very interesting area for research. However, most current research uses fixed channel assumptions and single-agent reinforcement learning methods, which do not account for the changing nature of real network channels and overlook the benefits of using multiple agents. To solve these problems, we suggest a new multi-agent reinforcement learning (MARL) system that can adjust offloading choices and resource distribution in NOMA-MEC networks as network conditions change over time. The proposed method employs quadratic transformation and convex approximation for efficient resource allocation and adopts a multi-agent MAPPO algorithm to address high-dimensional and dynamic environments. Many tests show that our new combined method uses much less energy than single-agent algorithms, and the theory behind it is supported by the results we got from experiments. Additionally, the strategy shows great flexibility and reliability in changing situations, highlighting its promise for smart resource management in future 6G-enabled MEC systems.