A user-based K-means clustering offloading algorithm for heterogeneous network

Abstract

Heterogeneous Networks (HetNets) is considered as a promising solution for 1000× mobile data challenge in the next decade. However, the different transmission power among various devices has caused load unbalance issues; the large scale small cells also require self-organization (SON) network to reduce computing complexity. K-means Clustering Algorithm (KCA) is a popular unsupervised Machine Learning (ML) algorithm which is used to solve classification problem. This paper will discuss the benefit and limitation of applying KCA to solve the two HetNets challenges, and then propose a User-Based K-means Algorithm (UBKCA) by involving HetNets background and Enhanced Inter Cell Interference Coordination (eICIC). Firstly, center user group set is established to reduce computing complexity. Secondly, CRE bias and Edge User Factor are introduced to enhance user offloading so that loading balance objective can be achieved. Simulations are then taken to show UBKCA's better performance than KCA; the optimal combination of CRE bias and Edge User Factor are taken based on both accuracy and offloading factor; and finally decision boundary is calculated to implement a close-loop SON system.