Greener Cellular Networks Toward Smart Cities

By Ryoichi Shinkuma

Traditionally, the primary purpose of cellular networks was to enable widespread communication by phone or email. Such functionality has recently been extended to machines like sensors, and could increasingly include autonomous cars, robots, and drones in the future. This potential for extension means that cellular networks serve as the cornerstone of the “smart city” concept [6], which proposes that cities utilize advanced technology to more effectively manage their services.

Cellular networks impose severe economic and ecological problems in the smart city era due to their substantial energy consumption; in 2016 alone, cellular networks worldwide are estimated to have consumed as much as \(4\) to \(5 \times 10^{11}\) kilowatt hours (kWh) of energy [4]. Hence, even a 0.1 percent reduction in cellular network power consumption is equivalent to \(4\) to \(5 \times 10^8\) kWh — which would save about 50 million U.S. dollars. Future networks will likely consume even more energy, which will greatly increase greenhouse gas emissions and exacerbate global warming. To address this matter of both wallet and planet, we must make smart cities’ cellular networks greener.

Reducing the energy consumption of the base stations (BSs) is critical to the realization of green cellular networks. These stations handle an enormous number of connections for cellular network users, thereby dominantly consuming 70 percent of a system’s overall energy. Previous research has found that 80 percent of the energy consumption in mobile communication networks stems from BS operation [1]. As such, switching off a single BS could lower consumption to nearly 40 percent of its maximum. 

Selectively switching BSs into sleep mode is an effective means of saving energy in mobile communication networks, especially when the traffic load is relatively low. Artificial intelligence (AI) technology is capable of estimating the temporal change in mobile traffic because it can forecast future traffic from past information (i.e., traffic logs). However, traffic logs cannot be collected from all BSs if some are switched off, which complicates this approach.

A new BS-control technology determines which BSs should be switched off based on the importance of their traffic logs [2, 3, 5]. We estimate the importance score of a BS's traffic log in terms of its contribution to improvements in traffic prediction accuracy. If the importance score is low, then the priority of the BS is low as well and it will be switched off to save energy. The AI technology has a function called “feature importance extraction” that estimates the importance score of the traffic logs of each BS. In the field of AI, “feature” is a well-used term that denotes information that characterizes the data for a specific AI task. For instance, if an AI task classifies an animal in a photo as either a dog or cat, then traits like the position and shape of the eyes, nose, and ears all become features for the task. 

As another example, indoor/outdoor smart monitoring systems use three-dimensional (3D) image data from 3D image sensors—such as light detection and ranging, better known as LIDAR—to perform anomaly detection on public roads or in indoor facilities. Sending sensor data to the server that hosts the AI is time consuming because the volume of image data is typically huge when compared to the limitations of communication bandwidth. One possible solution is to refrain from sending the raw data and instead use another AI extract the data’s features at the sensor; the sensor can then send the feature data—which has a much smaller volume than the raw data—to the server. Since the feature data contains the semantic information, the server-side AI can perform anomaly detection even without the raw data. However, a tradeoff does exist between bit reduction and the amount of semantic information in the feature data. For instance, if the feature data’s semantic information only contains details about moving vehicles on the road, then the server-side AI might fail to detect an object that is falling from a building. The research question thus becomes how to optimize this tradeoff. 

For the cellular traffic prediction task in the context of smart cities, we must consider the spatiotemporal correlation of traffic across neighboring BSs as a feature. In our system, we determine the importance scores of traffic logs from different BSs, then use this information for BS control [2]. BSs with high importance scores stay active while those with low importance scores switch to sleep mode. We have utilized computer simulation with a dataset of real traffic logs to examine the performance of this new technology, and found that our system performs better than two benchmark schemes in terms of prediction accuracy and robustness against different BS sets. The implementation of the technology in regard to smart cities is ongoing.

---

References
[1] Han, F., Zhao, S., Zhang, L., & Wu, J. (2016). Survey of strategies for switching off base stations in heterogeneous networks for greener 5G systems. IEEE Access, 4, 4959-4973.
[2] Shinkuma, R., Kishi, N., Ota, K., Dong, M., Sato, T., & Oki, E. (2021). Smarter base station sleeping for greener cellular networks. IEEE Network, 35(6), 98-103.
[3] Shinkuma, R., Yamada, Y., Sato, T., & Oki, E. (2020). Flow control in SDN-Edge-Cloud cooperation system with machine learning. In 2020 IEEE 40th international conference on distributed computing systems (ICDCS) (pp. 1304-1309). Singapore: Institute of Electrical and Electronics Engineers.
[4] Van Heddeghem, W., Lambert, S., Lannoo, B., Colle, D., Pickavet, M., & Demeester, P. (2014). Trends in worldwide ICT electricity consumption from 2007 to 2012. Comput. Commun., 50, 64-76.
[5] Yamada, Y., Shinkuma, R., Sato, T., & Oki, E. (2018). Feature-selection based data prioritization in mobile traffic prediction using machine learning. In 2018 IEEE global communications conference (GLOBECOM) (pp. 1-6). Abu Dhabi, UAE: Institute of Electrical and Electronics Engineers.
[6] Zhang, S., Zhao, S., Yuan, M., Zeng, J., Yao, J., Lyu, M.R., & King, I. (2017). Traffic prediction based power saving in cellular networks: A machine learning method. In Proceedings of the 25th ACM SIGSPATIAL international conference on advances in geographic information systems (pp. 1-10). Redondo Beach, CA: Association for Computing Machinery.

Ryoichi Shinkuma received his Ph.D. from Osaka University and is a professor at Shibaura Institute of Technology, prior to which he was an associate professor at Kyoto University and a visiting scholar in Rutgers University’s Wireless Information Network Laboratory. He is a senior member of the Institute of Electrical and Electronics Engineers and a Fellow of the Institute of Electronics, Information and Communication Engineers.