Optimal Cell Clustering and Activation for Energy Saving in Load-Coupled Wireless Networks

TL;DR

This paper develops an optimization framework for energy-efficient cell clustering and scheduling in load-coupled wireless networks, achieving over 60% energy savings compared to existing methods.

Contribution

It introduces a novel mathematical formulation, proves NP-hardness, and proposes algorithms including column generation and local enumeration for near-optimal solutions.

Findings

Achieves over 60% energy savings compared to existing schemes.

Solutions are within a few percent of the global optimum.

Provides a structural characterization of the optimization problem.

Abstract

Optimizing activation and deactivation of base station transmissions provides an instrument for improving energy efficiency in cellular networks. In this paper, we study optimal cell clustering and scheduling of activation duration for each cluster, with the objective of minimizing the sum energy, subject to a time constraint of delivering the users' traffic demand. The cells within a cluster are simultaneously in transmission and napping modes, with cluster activation and deactivation, respectively. Our optimization framework accounts for the coupling relation among cells due to the mutual interference. Thus, the users' achievable rates in a cell depend on the cluster composition. On the theoretical side, we provide mathematical formulation and structural characterization for the energy-efficient cell clustering and scheduling optimization problem, and prove its NP hardness. On the algorithmic side, we first show how column generation facilitates problem solving, and then present our notion of local enumeration as a flexible and effective means for dealing with the trade-off between optimality and the combinatorial nature of cluster formation, as well as for the purpose of gauging the deviation from optimality. Numerical results demonstrate that our solutions achieve more than 60% energy saving over existing schemes, and that the solutions we obtain are within a few percent of deviation from global optimum.