2327-4662 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2018.2871469, IEEE Internet of Things Journal 1 SensorRank: An Energy Efficient Sensor Activation Algorithm for Sensor Data Fusion in Wireless Networks Nashreen Nesa and Indrajit Banerjee Department of Information Technology Indian Institute of Engineering Science and Technology, Shibpur, Howrah, West Bengal 711103, India Email: {nashreennesa.rs2016, ibanerjee}@it.iiests.ac.in Abstract—Energy conservation is a challenging concern with the growing advancement in Internet of Things (IoT) technologies and is currently one of the attractive issues among researchers working in the IoT domain. In this paper, we address this issue by devising a dynamic sensor activation algorithm inspired by the popular PageRank algorithm called SensorRank. Unlike PageRank algorithm which only requires the outlinks to rank webpages, SensorRank, on the other hand, dynamically analyses the network topology in terms of relative distances and link qualities between devices and the remaining energies of the devices, based on which the sensors are ranked. An optimal subset of sensors are activated that take part in data fusion and the inactivated sensor data are regenerated with the help of Compressing Sensing technique. A quantitative study in terms of average energy gain is performed which shows an energy saving of upto 42% and a lifetime enhancement of upto 89% achieved through our proposed approach. Furthermore, comparison of SensorRank with random activation of sensors reveal that there is an overall lifetime enhancement of 34.68% with our proposed activation scheme. Index Terms—IoT, sensor activation, data fusion, PageRank, energy I. I NTRODUCTION Internet of Things (IoT) technologies have gone a long way in reducing the workload of humans by providing interconnec- tivity among all “things” imaginable such that everyday task is accomplished seamlessly and without manual intervention. This growing trend for developing IoT applications have resulted in a concern for energy efficiency of the deployed battery-powered IoT devices. Energy is a critical resource for maintaining the operation of these devices and the lifetime of the network inevitably depends on the energy of its individ- ual devices. Any device is expected to malfunction or stop functioning at all when the battery of the device is exhausted, which will result in an inconsistent behaviour of the entire application. Since IoT devices generate too much of redundant data for an end-user to process, a wise approach is to limit the amount of data to be sent for processing. This process of accumulating information is known as data fusion and it can help achieve large energy gains by combining data measurements from the most important and reliable sensor devices in the network. Obviously, activating a large number of devices in any network will result in better regeneration accuracy, but at the same time incurs huge energy cost by the activated devices. Thus, a sensible approach is to select the fewest essential subset of devices that also guarantees a reasonably acceptable regeneration accuracy. Researchers in the past have investigated the energy efficiency of power constrained in IoT nodes for enhancing the battery life of the deployed devices [1]–[3]. However, these are targeted to the development of sophisticated hardware for coping with energy efficiency issue. Recently, Compressive Sensing (CS) has attracted a lot of researches for developing energy-efficient data reconstruction algorithms [4]–[6]. CS is a technique of sampling in which any data/signal can be efficiently recovered with much fewer samples than Nyquist theory [7], [8]. It exploits some inherent characteristics (such as sparsity) of the signal or data in hand to compress it in a way so that only a subset of the entire data space is sent at the sink. The sink, on the other hand, will reconstruct the original data from the compressed version of the data with high probability. This will automatically lead to increase in the lifetime of the system. To further extend the lifetime of the network, few works [9], [10] in the literature adopts active-sleep cycle in which only a designated number of sensors will be active in a round. In this paper, we have devised a sensor activation algorithm called SensorRank that is inspired by the PageRank algorithm that Google uses to ascertain the importance of a web page dynamically. The intuition behind adapting PageRank in sensor networks is based on the similarity of the graphical structures of sensor networks and web graphs and the simple yet sophisticated approach of calculating their node importance. Specifically, our contributions are as follows : 1) We propose SensorRank, a dynamic sensor activation algorithm that prioritize the devices according to their remaining energies, relative distances and link qualities. 2) A subset of the most useful and essential devices are activated at the beginning of each round dynamically. 3) The data from the inactivated sensors are regenerated through the CS reconstruction algorithm with high prob- ability. 4) Extensive simulations were conducted with a real de- ployment of sensors in Intel laboratory and the per- formance of SensorRank were evaluated on the overall