Wireless Channel Characterisation for Low Power IoT Devices using Machine Learning Techniques

In recent years, wireless channel estimation has become crucial to ensure reliable communication in Low-Power Internet of Things (LP-IoT) networks. In these LP-IoT networks, the wireless channel constantly varies due to environmental factors such as noise, scattering, and multi-path fading. This fluctuation of the channel can have a severe impact on smooth communication. Therefore, there is a need to estimate the channel before accurate transmission and in result the effective information dissemination. Effective channel
estimation is essential for optimising network performance, especially in dynamic indoor environments. Numerous channel estimation techniques, such as have been applied to estimate cellular communication channels; however, channel estimation in low-power IoT devices is still being investigated using conventional and advanced techniques. The conventional estimation methods are typically designed without considering the unique constraints of LP-IoT devices and often struggle to capture the complex pattern of wireless channels in an indoor environment. These challenges lead to designing more effective and robust channel estimation solutions to achieve reliable communication in real-world applications. Furthermore, Machine Learning (ML) is increasingly being identified as one of the technologies utilised for future predictions in various fields, including wireless channel estimation. Traditional techniques also often rely on static models that fail to adapt to rapid changes in indoor environments, such as interference from physical obstructions, moving objects, or changes in signal reflection and absorption. However, ML models can dynamically learn from data patterns, adjusting their predictions in real time to reflect these
changing conditions, enhancing estimation accuracy. Furthermore, conventional techniques often rely on linear models that assume simple relationships between parameters, which may not hold in complex environments. ML algorithms, particularly Deep Learning (DL) models, excel at capturing nonlinear interactions, leading to more accurate estimations even in complex and irregular environments with nonlinear signal behaviours. Traditional methods are typically optimised for specific network setups and can only scale across varying IoT environments with extensive reconfiguration. In contrast, ML approaches can be generalised across different devices, networks, and environments by learning
patterns from diverse data, making them more flexible and easier to deploy on a large-scale smart city application. By overcoming the challenges of traditional estimation methods, ML-based channel estimation techniques can significantly enhance the robustness, accuracy, and adaptability of LP-IoT networks, making them well-suited for real-time applications in dynamic environments.
The research presented in this thesis is focused on estimating the wireless channel in LP-IoT communication by applying conventional and ML-based estimation techniques. An experimental test-bed was created to simulate a smart city environment using an interconnection of LP-IoT devices within a controlled indoor setting in the Network Engineering Laboratory at Charles Sturt University’s Port Macquarie campus to collect real data. A review of traditional wireless channel estimation techniques was conducted, which led
to the development of an advanced traditional estimation methods. Subsequently, ML-based estimation techniques were applied using Deep Learning and Random Forest models. For ML-based implementation, the research first established the method of primary scenarios using Artificial Neural Networks (ANNs), followed by developing advanced ANN-based estimations. Considering the higher training time of these models, this thesis investigates lightweight aspects of these models to cater to the LP-IoT devices and develops a Random
Forest Regressor (RFR)–based estimation model. A thorough comparison of the complexity of the developed models was conducted by deploying them on LP-IoT devices in real-time. Moreover, probability and scalability analysis of the selected estimation models has been conducted as a part of this research.
By conducting a thorough investigation in the field of LP-IoT channel estimation, this thesis has made substantial contributions to developing LP-IoT channel estimation strategies by introducing novel methodologies and performing comprehensive analyses, demonstrating the enhanced accuracy, efficiency, and real-world applicability of the proposed models. These contributions pave the way for further research and development in smart city IoT networks, addressing the evolving demands of LP-IoT wireless communication systems.