Random Neural networks and LoRaWAN for Location-based IoT Remote Health applications in smart cities

Abstract

The concept of the Internet of Things (IoT) has led to the interconnection of a significant number of devices and has impacted several applications in smart cities’ development. In the era of digital connectivity, information can be gathered, identified, controlled, and monitored with the help of wireless technologies. IoT has impacted various domains, including localization, healthcare, transportation, agriculture, and many more. In addition, Low-Power, Wide-Area Networks (LPWAN) are potential connectivity enablers within the IoT in smart cities. Long-Range, Wide-Area Network (LoRaWAN) is an open industrial, scientific, and medical radio band specification and an emerging LPWAN connectivity solution for IoT applications with minimum energy consumption. However, LoRaWAN is a relatively recent technology under current research and development. Localization is a vital IoT application that is being considered in different location-based IoT applications, such as in healthcare for the tracking of patients and the elderly, especially in case there is an emergency or in logistics and transportation for fleet management both in indoor and outdoor environments. Bluetooth, WiFi, and Zigbee wireless technologies have been successfully applied in indoor localization but are only limited to a few meters. Global Positioning Systems (GPS) is a frequently utilized solution for outdoor localization services; however, high-power consumption and hardware cost are significant hindrances to dense wireless sensor networks in large-scale urban areas. Also, GPS does work indoors but is limited due to a lack of direct Line-of-Sight (LOS) satellite signals from the End Device (ED) due to building materials blocking the ultra-high frequency signals. Various localization methods, including Received Signal Strength Indicator (RSSI) fingerprint localization techniques, are present in the literature, but all with different limitations. Therefore, LoRaWAN is being investigated in different location-based IoT applications due to more advantages with iv low-cost, long-range, and low-power characteristics. Furthermore, Random Neural Networks (RNN) has been used to develop different robust models in different IoT applications with significant accuracy, yet RNN applied to localization and heart disease prediction IoT applications is a research topic to be explored. In this thesis, a novel low-power intelligent localization model using LoRaWAN RSSI values is developed and evaluated using RNN for outdoor and indoor environments primarily to be used in location-based IoT Remote Health Monitoring (RHM) applications in smart cities where patients can be accurately located. Different RNN architectures and learning rates are used to evaluate the performance of the proposed LoRaWAN-RNN-based ED localization model using real-world RSSI LoRaWAN measurements collected in a three-floor building to evaluate the performance of the proposed model in an indoor scenario. Furthermore, real-world data sets collected from an outdoor urban area of Glasgow city and a dense urban area of Antwerp city are used to evaluate the performance of the developed model in an outdoor urban environment. The developed system is trained and used to predict unknown positions. It demonstrates a significantly higher accuracy compared to the results from the related work with a minimum mean localization error of 0.29 m in an outdoor urban environment and 0.12 m in LOS, and 13.94 m in Non-Line of Sight (NLOS) in indoor scenarios, respectively. The proposed localization system demonstrates 95% and 98.5% improvement in average localization error compared to related studies in outdoor and indoor environments, respectively. The obtained results confirm the developed LoRaWAN-RNN-based localization systems suitable for indoor environments in LOS, which can be applied in big sports halls, hospital wards, shopping malls, airports, and many more. Furthermore, the obtained results in NLOS confirm the developed systems appropriate for indoor NLOS applications, which can be applied in managing and controlling vehicles in indoor car parks, industries, or any other fleet in a pre-defined area.