Artificial neural network aided dynamic scheduling for eICIC in LTE HetNets

Abstract

In heterogeneous networks (HetNets), the deployment of small cells is accompanied by load balance and interference problems. A time-domain solution is enhanced Inter-Cell Interference Coordination (eICIC), which involves two parameters: Cell Range Extension (CRE) bias and Almost Blank Subframe (ABS) ratio. The centralized algorithm of jointly optimizing these two parameters is time consuming and is not suitable for HetNets with user mobility. In this paper, we propose an artificial neural network (ANN) aided scheduling scheme to enable fast reconfiguration in dynamic HetNets. An ANN is trained to learn the unknown relationship between the environment and the optimal CRE and ABS patterns. After training, the CRE and ABS patterns can be predicted directly instead of implementing a high-complexity algorithm. With the obtained eICIC parameters, the remaining problem is only single cell resource allocation, which can be solved easily by individual macro and pico. We analyze the performance in terms of utility and throughput of the proposed scheme in simulations.