Learning Backoff: Deep Reinforcement Learning-Based Wireless Channel Access

Taegyeom Lee

doi:10.1109/JSYST.2023.3309977

Learning Backoff: Deep Reinforcement Learning-Based Wireless Channel Access

2024-03-01

0

PRE

AI

DOI：10.1109/JSYST.2023.3309977

Original

Original request for help

Save

Abstract

En 中文

Unmanned aerial vehicles (UAVs) have been attracting a lot of interest for their significant advantages in terms of mobility and cost. Herein, we address a learning-based channel access method for UAV networks, which is named learning backoff algorithm. The proposed algorithm allows UAVs to determine the adequate backoff time without any information exchanges in distributed network environments. Each UAV in the network learns the network characteristics by analyzing its own location information and the pattern of the network obtained from the received signals. By employing the reinforcement learning (RL) model, UAVs learn repetitive flight patterns and select a proper backoff time to ensure link stability while avoiding collisions. The RL model utilizes a deep learning model and replay memory to handle the large amount of state-action data generated in UAV networks. The proposed algorithm shows promising results as a solution to the channel access problem in UAV networks.

Keywords:

Autonomous aerial vehicles

Throughput

Load modeling

Deep learning

Trajectory

Q-learning

Optimization

Autoscheduling

deep reinforcement learning (RL)

multiple channel access

unmanned aerial vehicle (UAV)

Learning Backoff: Deep Reinforcement Learning-Based Wireless Channel Access

Abstract

Journal

Organization