Home // ICN 2022, The Twenty-First International Conference on Networks // View article

Joint Power Control, Pilot Assignment, User Association and Flight Control for Massive MIMO Self-Organizing Drones using Reinforcement Learning

Authors:
Gabriel Skidmore

Keywords: Wireless Drone Networking; Massive MIMO; Deep Q-Learning; Reinforcement Learning; Nonconvex Optimization

Abstract:
Improving spectral efficiency is becoming increasingly important in mobile communications to keep up with the ever-increasing amount of data traffic coming from video streaming, Internet of Things, intelligent transportation systems, and augmented and virtual reality. In this work, a deep reinforcement learning algorithm (Deep Q-Learning) is implemented to maximize the sum spectral efficiency of ground users using Unmanned Aerial Vehicles (UAVs) as agents. The agents and environment are created by using OpenAI’s Gym library to create a custom implementation of the agent, reward function, and environment. The problem is then relaxed by assigning users to UAVs that lead to the highest Single-Input Single-Output (SISO) Signal to Interference plus Noise Ratio (SINR) and allowing the UAVs to assign multiple pilot signals to ground users. Lastly, the implementation of the algorithm is compared to a convex relaxed version of the original reward function.

Pages: 22 to 27

Copyright: Copyright (c) IARIA, 2022

Publication date: April 24, 2022

Published in: conference

ISSN: 2308-4413

ISBN: 978-1-61208-940-9

Location: Barcelona, Spain

Dates: from April 24, 2022 to April 28, 2022