Efficient joint resource allocation using self organized map based Deep Reinforcement Learning for cybertwin enabled 6G networks

Abstract

Abstract Sixth-generation wireless communication has emerged, stimulating the rapid growth of numerous types of real-time applications that are characterized by their high data computing demands and formation of massive data traffic. Cybertwin-enabled edge computing has become a logical way to satisfy the enormous user demands. However, there are drawbacks to this advancement as well. The effective distribution of resources while balancing the demands for computing, communication, and caching is a major problem in edge networks. The resource allocation problem in dynamic edge computing systems is too complex to address with traditional statistical optimization techniques. Therefore, a Joint Resource allocation method using Self-Organized Map (SOM)-based Deep Reinforcement Learning (DRL) is proposed for cybertwin-enabled 6G wired + wireless (hybrid) networks. This approach controls the clustering capabilities of SOM to organize the state space, followed by the decision-making strength of RL to select optimal actions for resource allocation in dynamic and real-time environments. The objective is to minimize overall latency and energy consumption. From the results analysis, using SOM-DRL, the hybrid network model outperforms the wireless-only model in terms of latency and energy consumption than the existing MATD3 method by achieves 3.34% of energy consumption, 3.17% of latency, and 7.30% of completion time.