Mobility aware and energy-efficient federated deep reinforcement learning assisted resource allocation for 5G-RAN slicing

Network slicing is one of the foundations for the realization of 5G and beyond. However, due to the mobility of the users and the network dynamics, flexible and efficient radio access network (RAN) resource slicing is still a challenge. In this paper, we allocate the power and radio block (RB) resources in a multi -RAN scenario to the users of both rate -based and resource -based slices. We propose a mobility -aware and energy -efficient federated deep reinforcement learning -assisted resource allocation (ME-FDRL-RA) method for RAN slicing in a large multiple RAN environment. ME-FDRL-RA includes both federated deep reinforcement learning (FDRL) and deep learning (DL) models as follows: Stacked and bidirectional long -short -term -memory (SBiLSTM), allocate resources to slices on a large time -scale. Additionally, federated advantage actor -critic (F-A2C) allocates resources on a small time -scale and speeds up convergence. Moreover, to solve the optimization problem of determining the required resources for the users of slices, we propose an efficient iterative algorithm called the interference -aware and energy -efficient power allocation (IA -EPA) method. According to simulation results, ME-FDRL-RA outperforms competitive methods in terms of convergence speed, computational complexity, energy efficiency, and the number of accepted users while addressing the challenges of user mobility and maintaining a desirable degree of inter -slice isolation.