An Effective Handover Scheme in Heterogeneous Networks using Multi Armed Bandit Based Learning Approach

Abstract

Deploying pico cell and femto cell nodes within a macro cell layout is known as heterogeneous networks. It is a promising solution to enhance overall system performance, cell-edges coverage. Indeed this type of deployment leads to an improvement of spectral efficiency and achieves load balance by offloading macro cell traffic to low power nodes. Heterogeneous networks deployment incurs new technical challenges related to handover performance of user equipment, which will be impacted especially when high velocity user equipment’s traverse pico cells. To tackle this problem, reinforcement learning techniques; Multi Armed Bandit and Bayesian Multi Armed Bandit has been proposed. User equipment’s learn the best cell based on the posterior distribution of reward and continuous optimal cell range expansion value is predicted through linear regression. These equipment’s are scheduled based on their velocity and previous rates (exchange among tiers). Information entropy is also used to evict the user equipment from overcrowded cell to the cell that has relatively less traffic, better throughput and higher signal to interference noise ratio. The potential reward on each base stations channel is calculated; then the channel with the maximum accumulated rewards is formally chosen. The proposed learning based approach with entropy measures for load balancing out performs the Multi Armed Bandit based mobility management in terms of user equipment throughput. In average, a gain of up to 86 % is achieved for user equipment throughput, while the handover failure probability is reduced to a factor of two by the proposed reinforcement based mobility management approaches. Simulation value of user equipment’s throughput validates the proposed scheme is better over the classical RSRP handover scheme.