Energy efficiency enhancement in heterogeneous networks using deep q-learning

Abstract

In the various researches that have been conducted about Heterogeneous Networks (HetNets), cell switching has been identified as one of the crucial strategies for reducing energy consumption. The major goal of this method is to turn off or offload traffic from idle or lightly loaded Base Stations (BSs) that surround an active Macro Base Station (BS). However, the major implication of offloaded traffic on surrounding MBS in terms of power consumption has not been adequately investigated. As a result, this thesis explores a Deep Q-learning (DQL) assisted cell switching algorithm to determine which Small Cells (SCs) should be turned off while taking into account the rise in the MBS’s power consumption due to offloaded traffic from the sleeping SCs. Another algorithm was also devised that focused on the switching off of the least loaded SCs. The research has fully explored a comparative study to justify the feasibility and the relevance of the proposed algorithm. Furthermore, the research adopted vertical offloading technique, where traffic was being offloaded from the SCs to the MBS. The research has mainly focused on the energy saving, capacity, power gain, and the energy efficiency that can result from using the proposed algorithm. From the simulations, the proposed algorithm has proved efficient in terms of both energy and power especially in the low traffic load regions. Simulations revealed that almost all the SCs can be switched off compared to the least loaded algorithm hence saving vast amount of power. For example, at 10% resource availability at the MBS, the proposed algorithm achieves a power gain of up to 27% compared to 7% attained with the least loaded scheme.