Learning annealing schedule of log-linear algorithms for load balancing in HetNets

Load balancing among the base stations in heterogeneous networks (HetNets) is essential for their successful deployment. In this paper, we present a robust approach for load balancing by adapting log-linear learning algorithms (LLLA). A new distributed annealing learning algorithm (ALA) is proposed to learn the parameter of LLLA by adapting successive reject algorithm. ALA gives a new annealing schedule that describes the evolution of parameter t of LLLA over a fixed horizon. The performance of this new annealing schedule is compared with commonly used annealing schedules in the literature such as linear decreasing, log decreasing, and fixed parameter. It is observed from simulations that the new annealing schedule achieves lower global cost for a fixed time horizon compared to that of other annealing schedules. For lower time horizons, ALA with linearly decreasing t is better than ALA with a fixed vector of t. Whereas, for higher time horizons, ALA performance is same in both the cases. Finally, we show the applicability of the proposed algorithm for load balancing.