Uplink Multiuser Scheduling Using Machine Learning

Abstract

Multiuser scheduling enables users to share the same time and frequency resources while exploiting spatial diversity through the use of multiple antennas. In this paper, we propose a machine learning (ML) approach that decides on multiuser scheduling through solving a system capacity optimization problem. More specifically, we use a support vector machine (SVM). The proposed algorithm takes as an input the signal to noise ratio (SNR) and uplink channel information of a predetermined set of users. The output is a decision as to which users, if any, can be scheduled in the same time slot and frequency band. We show that the resulting system capacity is comparable to the optimal capacity obtained through exhaustive search, with significantly lower algorithm complexity. Moreover, building on the crucial importance of feature-engineering in ML models and capitalizing on the domain-expert knowledge of our problem, we work on tailoring the information available at the scheduler to further enhance the performance of our proposed approach.