Collaborative Artificial Intelligence (AI) for User-Cell association in Ultra-Dense Cellular Systems

TL;DR

This paper introduces a collaborative AI approach using neural Q-learning for user-cell association in ultra-dense wireless networks, enabling faster convergence and improved data rates without extensive data exchange.

Contribution

It proposes a novel imitation-based neural Q-learning algorithm for cell association, leveraging local and neighboring data to enhance performance in dense networks.

Findings

Faster convergence to optimal cell association.

Improved data rates compared to traditional methods.

Effective collaboration without direct data exchange.

Abstract

In this paper, the problem of cell association between small base stations (SBSs) and users in dense wireless networks is studied using artificial intelligence (AI) techniques. The problem is formulated as a mean-field game in which the users' goal is to maximize their data rate by exploiting local data and the data available at neighboring users via an imitation process. Such a collaborative learning process prevents the users from exchanging their data directly via the cellular network's limited backhaul links and, thus, allows them to improve their cell association policy collaboratively with minimum computing. To solve this problem, a neural Q-learning learning algorithm is proposed that enables the users to predict their reward function using a neural network whose input is the SBSs selected by neighboring users and the local data of the considered user. Simulation results show that the proposed imitation-based mechanism for cell association converges faster to the optimal solution, compared with conventional cell association mechanisms without imitation.