Distributed Drone Base Station Positioning for Emergency Cellular Networks Using Reinforcement Learning

Paulo V Klaine,Muhammad A Imran,João P B Nadas,Richard D Souza

doi:10.1007/s12559-018-9559-8

Paulo V Klaine, Muhammad A Imran + Show 2 more

Open Access

https://doi.org/10.1007/s12559-018-9559-8

Copy DOI

Abstract

Due to the unpredictability of natural disasters, whenever a catastrophe happens, it is vital that not only emergency rescue teams are prepared, but also that there is a functional communication network infrastructure. Hence, in order to prevent additional losses of human lives, it is crucial that network operators are able to deploy an emergency infrastructure as fast as possible. In this sense, the deployment of an intelligent, mobile, and adaptable network, through the usage of drones—unmanned aerial vehicles—is being considered as one possible alternative for emergency situations. In this paper, an intelligent solution based on reinforcement learning is proposed in order to find the best position of multiple drone small cells (DSCs) in an emergency scenario. The proposed solution’s main goal is to maximize the amount of users covered by the system, while drones are limited by both backhaul and radio access network constraints. Results show that the proposed Q-learning solution largely outperforms all other approaches with respect to all metrics considered. Hence, intelligent DSCs are considered a good alternative in order to enable the rapid and efficient deployment of an emergency communication network.

Highlights

Sporadic, natural large-scale disasters, such as earthquakes, hurricanes, and tsunamis, produce a profound impact in human society, in terms of the infrastructure that is destroyed, but most importantly, in terms of human lives that are lost
The proposed solution is compared to different positioning strategies, such as deploying the drones in fixed random positions, fixed around a circle centered in the middle of the area at evenly spread angles, and deploying the drones in the locations of hot spots of the previous destroyed network, and the results show that the intelligent Q-learning solution outperforms all of them in all considered metrics
Since it might be difficult for the operator to deploy the macro base station (BS) in its original position, either due to debris or blockages, in the simulations, the truck containing the macro BS is positioned at its initial position added to an offset depending on a random distribution

Summary

Introduction

Natural large-scale disasters, such as earthquakes, hurricanes, and tsunamis, produce a profound impact in human society, in terms of the infrastructure that is destroyed, but most importantly, in terms of human lives that are lost. More robust solutions, involving networks that are capable of self-organization and that can be deployed quickly and effectively to the exact area where coverage is needed, Cogn Comput (2018) 10:790–804 should be designed. In this sense, algorithms that can adapt themselves, such as artificial intelligence and machine learning, should be deployed [1,2,3,4], to enable a fully autonomous network

Objectives

Methods

Results

Conclusion