A Machine Learning Based Cell Switch-off Algorithm In Cellular System

With the increasing usage of wireless devices, the power consumption and energy efficiency of cellular networks have been receiving much attention in wireless network in recent years. One way of saving energy is to switch off some underutilized small cells in light traffic conditions. In existing researches, the number of the base stations (BSs) that can be switched off in a network is take as the optimization object. However,under many circumstances, it would be acceptable if the energy efficiency increases with less cells switched off. In this paper, we take the energy efficiency as the optimization object and suppose that each user equipment(UE) has a minimum rate requirement and the bandwidth is uniformly shared by UEs in a cell. Under this assumption, the problem of optimizing cell switch-off for maximum energy efficiency is still NP-hard.We present a machine learning based small cell switch-off algorithm in omnidirectional cell deployment. The machine learning based small cell switch-off algorithm consists of two phases. The first phase is to get the initial switch-off pattern by an annual neural network based algorithm. And the second phase is to get the best switch-off pattern by optimization algorithms. We use two optimization algorithms in this phase, those are a Hamming distance based search algorithm and a simulated annealing based optimization algorithm. Numerical results are presented through system level simulations. The results showed that the algorithms in this paper outperform the benchmark of the random search method with relatively low cost in the engineering solution.