Convolution Neural Network Based Dynamic Pico Cell Operation for Multi-tier Heterogeneous Networks

Abstract

We consider a three-tier heterogeneous network (HetNet), consisting of macro-cell base stations (MBSs), small-cell base stations (SBSs) and pico-cell base stations (PBSs), serving both Internet of things (IoT) devices with low data rate requirements and general user equipments (UEs). In order to reduce the power consumption caused by dense deployment of base stations (BSs), we propose a novel machine-learning (ML) based dynamic PBS operation scheme. In contrast to conventional PBS on/off operation scheme, the proposed ML-based dynamic PBS operation scheme can dynamically change the PBS on/off status according to UEs' real-time location so as to reduce the total power of BSs. Specifically, we use the convolution neural network (CNN) algorithm to solve the proposed optimization problem. Simulation results show that CNN algorithm can achieve 68% performance of the optimum level in terms of power consumption saving while the calculate complexity is $\mathcal{O}(1)$ .