A Low Complexity ML-Assisted Multi-knapsack-based Approach for User Association and Resource Allocation in 5G HetNets

Abstract

Small cells are being deployed in the most recent Heterogeneous Network (HetNet) environments, which are supported by 5th-generation (5G) network solutions. They improve the performance of conventional macro-cell networks. Before data transmission in HetNets begins, a process of user association (UA) with a particular base station (BS) is triggered. Diverse resource allocation (RA) algorithms are also used during data transmission. Enhancing network load balancing, spectrum performance, and energy efficiency are all important goals that UA-RA solutions help to achieve. The authors of this paper previously presented a meta-heuristics algorithm for solving UA-RA problems in HetNets known as Performance-Improved Reduced Search Space Simulated Annealing (PIRS3A). Still, PIRS3A is only suitable for scenarios with low number of users as its complexity increases with the increase in the number of users. In this paper, we enhance the PIRS3A to support a large number of users by using Machine Learning (ML). We propose the Enhanced PIRS3A (EPIRS3A) which uses supervised ML techniques to classify users into two separate classes, i.e., users who need offloading and users who do not. PIRS3A will be used to solve the UA-RA problem for those users who need offloading in order to improve the quality of their services.