Reinforcement Learning Empowered Unmanned Aerial Vehicle Assisted Internet of Things Networks
View/ Open
Metadata
Show full item recordAbstract
This thesis aims towards performance enhancement for unmanned aerial vehicles (UAVs) assisted internet of things network (IoT). In this realm, novel reinforcement learning (RL) frameworks have been proposed for solving intricate joint optimisation scenarios. These scenarios include, uplink, downlink and combined. The multi-access technique utilised is non-orthogonal multiple access (NOMA), as key enabler in this regime. The outcomes of this research entail, enhancement in key performance metrics, such as sum-rate, energy efficiency and consequent reduction in outage. For the scenarios involving uplink transmissions by IoT devices, adaptive and tandem rein forcement learning frameworks have been developed. The aim is to maximise capacity over fixed UAV trajectory. The adaptive framework is utilised in a scenario wherein channel suitability is ascertained for uplink transmissions utilising a fixed clustering regime in NOMA. Tandem framework is utilised in a scenario wherein multiple-channel resource suitability is ascertained along with, power allocation, dynamic clustering and IoT node associations to NOMA clusters and channels. In scenarios involving downlink transmission to IoT devices, an ensemble RL (ERL) frame work is proposed for sum-rate enhancement over fixed UAV trajectory. For dynamic UAV trajec tory, hybrid decision framework (HDF) is proposed for energy efficiency optimisation. Downlink transmission power and bandwidth is managed for NOMA transmissions over fixed and dynamic UAV trajectories, facilitating IoT networks. In UAV enabled relaying scenario, for control system plants and their respective remotely deployed sensors, a head start reinforcement learning framework based on deep learning is de veloped and implemented. NOMA is invoked, in both uplink and downlink transmissions for IoT network. Dynamic NOMA clustering, power management and nodes association along with UAV height control is jointly managed. The primary aim is the, enhancement of net sum-rate and its subsequent manifestation in facilitating the IoT assisted use case. The simulation results relating to aforesaid scenarios indicate, enhanced sum-rate, energy efficiency and reduced outage for UAV-assisted IoT networks. The proposed RL frameworks surpass in performance in comparison to existing frameworks as benchmarks for the same sce narios. The simulation platforms utilised are MATLAB and Python, for network modeling, RL framework design and validation.
Authors
Mahmud, SKCollections
- Theses [4186]