User association-based load balancing using reinforcement learning in 5G heterogeneous networks

The deployment of heterogeneous networks (HetNets), which include macro and micro Base Stations (BSs), has emerged as an effective solution to address the challenges faced by 5G communication networks. HetNets leverage the strengths of various technologies to enhance coverage, capacity, quality of service (QoS), cost efficiency, and adaptability. Efficient methods for associating user equipment with BSs in HetNets are necessary to maximize overall network performance. This process, known as user association-based load balancing, is crucial due to the increasing number of devices requiring connectivity using limited available spectrum and BSs. This study considers a three-tier downlink HetNet consisting of macro-BSs, pico-BSs, and femto-BSs, operating under a log-distance channel model in both rural and urban areas, in accordance with 3GPP norms. The signal model for the proposed HetNet is devised to determine the signal-to-interference-plus-noise ratio (SINR) and transmission rates for each user association. Load balancing is then performed using the deep Q-learning approach, and the cumulative transmission rate achieved is calculated and tabulated. This approach is compared with the MaxSINR and Q-Learning algorithms. LB (load balancing) was performed for different network scenarios by varying the number of UEs and BSs and introducing mobility in the network. The corresponding results were analyzed and plotted. An extensive survey of the effect of user association LB on improving the cumulative transmission rate of HetNets was conducted successfully, and the variation in performance under different network scenarios was analyzed.