Degree Name

Doctor of Philosophy


School of Electrical, Computer and Telecommunications Engineering


An Internet of Things (IoT) network consists of multiple devices with sensor(s), and one or more access points or gateways. These devices monitor and sample targets, such as valuable assets, before transmitting their samples to an access point or the cloud for storage or/and analysis. A critical issue is that devices have limited energy, which constrains their operational lifetime. To this end, researchers have proposed various solutions to extend the lifetime of devices. A popular solution involves optimizing the duty cycle of devices; equivalently, the ratio of their active and inactive/sleep time. Another solution is to employ energy harvesting technologies. Specifically, devices rely on one or more energy sources such as wind, solar or Radio Frequency (RF) signals to power their operations. Apart from energy, another fundamental problem is the limited spectrum shared by devices. This means they must take turns to transmit to a gateway. Equivalently, they need a transmission schedule that determines when they transmit their samples to a gateway.

To this end, this thesis addresses three novel device/sensor selection problems. It first aims to determine the best devices to transmit in each time slot in an RF Energy-Harvesting Wireless Sensor Network (EH-WSN) in order to maximize throughput or sum-rate. Briefly, a Hybrid Access Point (HAP) is responsible for charging devices via downlink RF energy transfer. After that, the HAP selects a subset of devices to transmit their data. A key challenge is that the HAP has neither channel state information nor energy level information of device. In this respect, this thesis outlines two centralized algorithms that are based on cross-entropy optimization and Gibbs sampling.

Next, this thesis considers information freshness when selecting devices, where the HAP aims to minimize the average Age of Information (AoI) of samples from devices. Specifically, the HAP must select devices to sample and transmit frequently. Further, it must select devices without channel state information. To this end, this thesis outlines a decentralized Q-learning algorithm that allows the HAP to select devices according to their AoI.

Lastly, this thesis considers targets with time-varying states. As before, the aim is to determine the best set of devices to be active in each frame in order to monitor targets. However, the aim is to optimize a novel metric called the age of incorrect information. Further, devices cooperate with one another to monitor target(s). To choose the best set of devices and minimize the said metric, this thesis proposes two decentralized algorithms, i.e., a decentralized Q-learning algorithm and a novel state space free learning algorithm. Different from the decentralized Q-learning algorithm, the state space free learning algorithm does not require devices to store Q-tables, which record the expected reward of actions taken by devices.

FoR codes (2008)



Unless otherwise indicated, the views expressed in this thesis are those of the author and do not necessarily represent the views of the University of Wollongong.