Energy savings under performance constraints via carrier shutdown with Bayesian learning

TL;DR

This paper introduces a Bayesian learning-based method for optimizing carrier shutdown thresholds in base stations to reduce power consumption while maintaining key performance indicators within acceptable limits, validated in a live 4G network.

Contribution

It presents a novel closed-loop Bayesian approach for dynamic carrier shutdown threshold optimization to balance power savings and performance.

Findings

11% reduction in power consumption at the base station

KPI targets were successfully maintained

Method validated in a real-world 4G network

Abstract

By shutting down frequency carriers, the power consumed by a base station can be considerably reduced. However, this typically comes with traffic performance degradation, as the congestion on the remaining active carriers is increased. We leverage a hysteresis carrier shutdown policy that attempts to keep the average traffic load on each sector within a certain min/max threshold pair. We propose a closed-loop Bayesian method optimizing such thresholds on a sector basis and aiming at minimizing the power consumed by the power amplifiers while maintaining the probability that KPI's are acceptable above a certain value. We tested our approach in a live customer 4G network. The power consumption at the base station was reduced by 11% and the selected KPI's met the predefined targets.