A unified spatiotemporal sleep mode approach for energy efficient dense HetNet’s using machine learning

Abstract

Future cellular networks are characterized by dense deployment of heterogeneous networks due to the ever increasing data traffic demand. However, the dense deployment of small base stations in a heterogeneous network presents high capital and operational expenditure for network operators. Dense deployment of small base stations not only increases the network's energy consumption but also increases the contribution of cellular networks to the global carbon footprint. Even though networks are designed to handle peak traffic, traffic demand is quite variable in both location and time, hence the need to adapt network energy consumption to these unavoidable variations in traffic. This research looks at how energy consumption can be adapted to the space-time traffic demand of the network by utilizing sleep mode schemes and machine learning. Machine learning is used to capture temporal variations in cellular network traffic and predict future trends of the data. A deep learning-based artificial neural network, the Long Short Term Memory (LSTM) network, is used for cellular traffic forecasting. The forecast data is eventually mapped to the network sum rate equivalents obtained from system-level simulations of the heterogeneous network. Various sleep mode approaches are then applied to the network to adapt power consumption to the temporal traffic segments of the forecast data. This research considers the spatial aspect of traffic by using location optimization to improve the performance of the sleep mode approaches. The performance of the location optimization-based sleep mode schemes is then compared with that of the conventional schemes. The location optimization-based strategic sleep mode approach, which prioritizes base stations that serve the most users, outperforms the conventional sleep mode strategies in terms of energy performance. At the peak hour traffic segment, the location optimization-based strategic sleep mode provides energy savings of up to 2 kW compared to the conventional sleep mode strategy which provides energy savings of not more than 700 W. The results obtained show that a machine learning-driven base station sleep mode approach that considers spatiotemporal traffic variations can be used to develop energy-efficient cellular networks that reduce both operational and capital expenditure by the network operators.