Cooperative and Reinforcement Learning in Energy Efficient Dual Hop Clustered Networks

Abstract

This paper examines the application of distributed Reinforcement Learning RL to improve the spectral efficiency in
high data rate applications for clustered networks. With RL, cluster members can learn to identify set of channels, which have the
highest success rate. It is shown that RL can minimize the dual hop clustered networks interference as the uplink delay is reduced
by up to 30% and improve the network energy efficiency of by up to 10% compared to a random channel allocation. However,
distributed RL has a very poor convergence time. In this paper, we present two methodologies on how through cooperative
learning and RL, cluster members can exchange channel historical information to facilitate learning. The proposed cooperative
learning methods enable cluster members to enter exploitation stage by a factor of 3 times faster compared to distribute RL.
Furthermore, the proposed methods allows each cluster to adapt the number of required preferred channel size depending upon
the local area density and traffic. The results shows that the adaptability reduces variation in uplink delay between cluster
members by 45% compared to distributed RL making the system more equal.