Abstract
Recently, the topic for the development of efficient handover algorithms to select the target (serving) base station (BS) such that the average throughput of user equipments (UEs) is maximized in heterogeneous cellular networks has received much attention. In the literature, the latest handover algorithms employ the upper confidence bound (UCB) policy, one efficient method for the multi-armed bandit (MAB) problem, since the UCB policy can identify the critical actions quickly by taking the uncertainty of the action reward into account. However, the UCB policy does not take advantage of the observed features, which could lead to poor performance in the UE throughput since the throughput of a UE served by a BS is highly correlated to the observed features including the bandwidth of the BS, the number of UEs served by the BS, and the spectral efficiency of the UE achieved by the BS. By contrast, besides the consideration of the uncertainty of the action reward, the linear upper confidence bound (LinUCB) policy also makes use of the observed features to estimate the action reward. In this paper, we make the first attempt to propose a LinUCB-based handover algorithm. The challenge is to define the reward and observed features highly correlated to the UE throughput such that the reward is a linear combination of the observed features as demanded by the LinUCB policy. We conduct simulations on the network simulator ns-3 to show that the proposed algorithm outperforms the state-of-the-art handover algorithms in terms of the average UE throughput.