HetNet Spatial Topology Design Using Mini-Batch K-means Clustering

In this paper we present framework for improving the heterogeneous network (HetNet) topology using unsupervised mini-batch online K-Means clustering algorithm. HetNet placement modelling often relies on the stochastic geometry processes, including Poisson point process (PPP) or Binomial point process (BPP) without incorporating a priori knowledge of user location. On the contrary, our proposal takes advantage of the available information regarding user mobility patterns. In order to mimic the complex dynamic users' mobility, we characterize the user's mobility using two different patterns: deterministic model based on Bézier curves, and stochastic Lévy flight mobility model. The simulation results confirmed our expectations and showed that refined HetNet topology provides higher throughput compared to that provided by BPP modelling approach.