4G Uplink Power Control Tuning Approach Using Unsupervised Machine Learning

Abstract

In this paper, we study a data driven approach using features selection and unsupervised machine learning techniques to analyze 4G cells behaviour from uplink performance perspective and group network cells into clusters that have commonalities in uplink behavior. Followed by the addressing of these commonalities with an appropriate values for the parameters controlling the uplink power control in each cluster, using domain knowledge in order to maximize the spectral efficiency and improve the performance. We leveraged one month of data collected from tier one mobile operator serving more than 30 million subscribers, and extensively investigated up to which extent the adopted approach efficiently optimize the 4G performance with better Quality of Experience (QoE) in comparison to a domain experience approach only. The adopted approach showed an improvement in uplink speed for live 4G network by 7%. Additionally, the proposed approach may be extended to tackle other optimization problems helping operators to accelerate machine learning adoption in solving the increasing challenges they are facing to best utilize network resources and improve performance.