Energy-Saving Small-Cell Wake Up Strategy for Ultra-Dense Networks Based on User Behavior Prediction

Ultra-dense networks (UDNs) is one of the key technologies that can offer large capacities with numerous small base stations (BSs) deployed. However, the BSs consume a lot of energy when they are all turned on. To save energy, small cells with zero or low load should be in sleep mode. In this paper, we propose a small cell wake-up strategy based on the mobile application usage prediction of the users. First, LSTM neural network model is used to predict the users’ application usage in the next interval and the training data is from a real-world anonymous datasets. The small BS will be woken up when it is predicted that one or more users in its coverage area will use high data-rate applications in the next time period. Numerical results show that the LSTM method achieves higher prediction precision and recall compared with the other prediction algorithms. Employing our scheme, we can get about 14% gain in energy consumption compared to the energy efficient system where the small cell is woken up based on the predicted number of users in its coverage area.