Satellite-based localization methods for Internet-of-Things applications: modeling approaches and performance evaluation

Internet-of-Things (IoT) is revolutionizing nearly all industries, as businesses are transforming their operation by linking the physical world with the digital realm. Examples of IoT applications can be found in smart cities, livestock monitoring, and logistics tracking. Many of these applications have a geo-relevance of the collected data. For example, in livestock tracking, it is important to reduce labor demand and increase efficiency in monitoring livestock grazing and strengthen livestock security. However, IoT access technologies have specific constraints, such as limited energy supply, computational resources, and communication overhead, it is challenging to localize these devices using conventional methods such as the Global Navigation Satellite System (GNSS). There are many types of IoT access technologies that are available to date, such as the Low Power Wide Area Network (LPWAN), cellular, Bluetooth, and Wi-Fi. LPWAN technologies such as LoRa, Narrowband (NB)-IoT and Sigfox are some of the main technologies for the wide-area IoT. While the GNSS is one of the best solutions in providing an accurate position of devices, it is deemed not always suitable to be applied in IoT context as the core characteristics of IoT technologies are low power and low cost. Thus, we look through other localization techniques that better satisfy the LPWAN requirements. Moreover, this research focuses on the non-terrestrial wide-area IoT due to recent developments of LEO satellites which have opened various opportunities for IoT-over-satellite applications. Therefore, we investigate the possibilities of using alternative localization methods and analyze their performance. Localization methods can be classified into range-based, which can be made in three main domains: power, time, and space (angle), and range-free such as hop count and pattern matching (fingerprint) technique. This research builds on clues provided by the received signal strength (RSS) fingerprint, Doppler fingerprint, and angle-of-arrival (AoA) methods. These measurements are readily available in the IoT gateway, thus simplifying the IoT localization hardware architecture. For RSS fingerprinting technique, it is crucial to address realistic spatial channel modeling as it is expected to affect the RSS pattern that is generated using the path-loss calculation. RSS pattern is modeled from Friss transmission with the addition of large-scale fading effect. The pattern is unique due to the differences in distance between the transmitter and receiver, the geometry of the location, and the randomness in large-scale fading at a specific location. Since each location exhibits unique pattern, inaccuracy in the RSS pattern will impact the localization performance. In satellite applications, Doppler pattern can be utilized for localization as the frequency shift due to the satellite’s relative velocity with respect to the ground device is distinctive to the device’s location. Also, AoA measurements are used in the localization using a satellite that is equipped with a multi-antenna. AoA provides an estimated direction of the ground device from the satellite perspective. Thus, in this research, by combining the Doppler pattern with the AoA, a higher localization accuracy is expected. Since we are dealing with pattern techniques for localization, we investigate the use of machine learning algorithms and statistical analysis such as Bayesian inference in determining the location from the measurements. Moreover, stochastic optimization methods are also considered and to be paired with the likelihood function derived from the statistical models.  By optimizing the unknown parameters (device’s location) with a goal to maximize the likelihood function, the localization of the device can be successfully performed. This research also models and optimizes the suitable localization techniques for IoT applications. We focus on the IoT-over-satellite applications as it is growing rapidly. The localization modeling is based on the realistic representation of the measurements to accurately translate the analysis to a real system deployment. Furthermore, we provide alternative localization methods in satellite-based localization as there is a huge expansion of IoT technologies using LEO satellites. The localization performance for the alternative localization methods is evaluated in terms of its accuracy which is the Euclidean distance between the ground truth and the estimated location. In RSS fingerprinting, radio shadowing is one of the contributors to the localization error. Shadowing is the random variations in RSS due to specific geometries of the paths between the transmitter and receiver. It is affected by objects such as trees, hills, and buildings. Typically, the random shadowing effect is modeled as uncorrelated. However, realistically, the shadowing is spatially correlated since the obstruction geometry does not change significantly around the neighboring areas. Therefore, we investigate the effect of introducing correlation in the shadowing, where we prove that the localization performance using machine learning algorithms improves as the correlation distance increases. For the satellite-based localization, we study the performance of RSS and Doppler pattern, and the AoA and Doppler pattern. In both studies, we derive the likelihood functions which represent the joint probability of the measurements given the location. From the likelihood, we then apply stochastic optimization methods to estimate the location that best reproduces the observed measurements. Since the measurements can be relayed to an IoT gateway, the localization can be performed on the gateway, thus reducing the computational and power load on IoT device compared to the GNSS solution which draw power from the IoT devices themselves. From the results, we observe that the localization performance using AoA measurements is better than RSS due to the smaller error in AoA measurements. RSS is more susceptible to a range of conditions that can degrade its signal such as distance and atmospheric effect, thus contributing to a larger deviation in the measurement. Here, we are more interested to utilize the stochastic optimization rather than machine learning algorithms in the framework. To cater for the random and short transmission of ground IoT devices, we develop a satellite visibility window determination based on Doppler measurements. This allows ground devices to only transmit during the satellite availability window to ensure sufficient measurements for localization are obtained and to preserve the devices battery. Lastly, we hope our work is beneficial to the IoT localization field, especially in the accelerating IoT-over-space applications, thus helping industry and research across the globe to further advance the field.