Learning Based Channel Access, Data Collection and Computation Methods for Energy Harvesting IoT Networks
Doctor of Philosophy
School of Electrical, Computer and Telecommunications Engineering
Internet of Things (IoT) networks consist of sensing devices and gateways. Specifically, these devices monitor an environment to obtain measurements of a physical quantity such as temperature or the location of a target. A gateway or server is then required to collect and compute sensory data from devices. A critical issue in an IoT network is that devices have limited operation time due to the lack of energy. This affects the amount of data collected by devices and data processed by a gateway. To this end, prior works have considered powering devices using energy sources such as solar or wirelessly via Radio Frequency (RF) signals. Another issue is ensuring sensed data is processed quickly to infer any events. To address this issue, many works have considered installing computational resources at the edge of an IoT network.
Given the above issues, this thesis first considers a Hybrid Access Point (HAP) that charges one or more energy harvesting devices via RF signals. These devices then transmit their data to the HAP. It considers Dynamic Framed Slotted Aloha (DFSA) protocol for channel access and devices can only transmit if they have sufficient energy. Moreover, nodes are not aware of channel state information (CSI) and each other’s energy level, meaning the HAP and devices are unaware of the number of devices that are ready to transmit. In addition, it considers different nonlinear energy conversion models. Given these considerations, this thesis outlines a two-layer approach, where at the first layer, the HAP adjusts its transmission power using a Sequential Monte Carlo (SMC) approach, and its frame size according to the Softmax function. At the second layer, devices use another Softmax function to learn the time slot that yields the highest reward for a given frame size. The results indicate that the two-layer learning approach achieves at least 7%, 19%, 40% higher throughput than Time Division Multiple Access (TDMA), i.e., each device is assigned to a dedicated data slot, ϵ-greedy and Aloha.
Yu, Hang, Learning Based Channel Access, Data Collection and Computation Methods for Energy Harvesting IoT Networks, Doctor of Philosophy thesis, School of Electrical, Computer and Telecommunications Engineering, University of Wollongong, 2022. https://ro.uow.edu.au/theses1/1516
FoR codes (2008)
080503 Networking and Communications
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.