Supervised-Learning-Based Resource Allocation in Wireless Networks

Abstract

A supervised-learning-based distributed resource allocation with limited information exchange is addressed for the proportional fairness maximization. In the future ultra dense networks, excessive network overhead is required for acquiring global channel state information (CSI). Hence, only partial CSI is generally available at each SBS. With partial CSI, however, it is almost impossible to perform optimal resource allocation without any heuristic problem relaxation or iterative information exchange. This is because the relationship between the proportional fairness and the partial CSI is unknown. In this paper, our aim is to design resource allocation considering the unknown relationship between partial CSI and proportional fairness by deep learning. For our proposal, we collect and construct dataset for supervised learning. In the numerical results, it is shown that the proposed scheme shows better performance than the conventional fair resource allocation scheme.