Deep Adaptive Reinforcement Learning‐Based Energy Efficient Clustering and Resource Allocation Mechanism in Ultra‐Dense Network

Abstract

ABSTRACT Due to the quick enhancement of the 5th Generation (5G) wireless communications, the user count is enhancing relatively. So, Ultra‐Dense Networks (UDNs) are becoming very significant for assisting various users and enhancing mission‐critical applications. To link numerous user systems and offer better data rates, the UDN is referred to as an effective innovation in 5G. It boosts the overall network functionality by enhancing the throughput of the system and managing the load balance of the system. But, the enhanced power utilization and the interference of the UDN because of the deeply installed Base Stations (BSs) are unavoidable. In order to tackle the interference and computation issues in the UDN, a clustering procedure is suggested. In the resource allocation process, the clustering procedure helps to tackle the computational issues by converting the entire cluster into sub‐clusters. But, these techniques face more issues in terms of users' Quality of Service (QoS), and they also lag in ensuring the system's transfer efficiency. Thus, to tackle these difficulties, new energy‐efficient clustering and resource allocation mechanisms for UDN have been introduced in this paper utilizing adaptive deep learning. First, an optimal clustering process is performed to choose the clusters optimally using Modified Hybrid Leader‐based Optimization (MHLO). Moreover, several multi‐objective functions are derived using MHLO and this process also supports managing the traffic load among the clusters. Subsequently, in order to enhance convergence and minimize computational complexity, resource allocation is performed in UDN. For this objective, Deep Adaptive Reinforcement Learning (DARL) is employed, where the network parameters are optimized by the MHLO algorithm. In the end, numerical and experimental research is carried out on the developed energy‐efficient clustering and routing model over diverse traditional algorithms and techniques. The developed framework MHLO‐DARL accomplished higher throughput by 3.4%, a success rate of 97%, and energy efficiency of 0.85%, which seems to be better than recent techniques like GAN‐DDQN, DDQN, STPD‐RL, and MDQN‐MAR. Hence, the validation outcomes showed that the suggested mechanism is more efficient in allocating the resources and executing clustering procedures in the complex environment without affecting communication in a network.