Spatiotemporal Function Estimation and Sensor Navigation

Many physical and technical parameters vary with both spatial position and time. Examples include temperature, air pressure, humidity, cancer mortality, magnetic field intensity, water salinity, pollutant concentration in a chemical cloud, incidence rate in an epidemic, quality of service in mobile communications, and many others. In this project, the main goal is the estimation of spatiotemporal functions describing position- and time-dependent parameters. This estimation uses measurements taken by remotely controlled sensors such as unmanned aerial vehicles (drones). Two additional goals of the project are sensor localization (estimating the varying positions of the sensors) and sensor control (controlling the movement of the sensors such that the accuracy of spatiotemporal function estimation is maximized). The new approach of the project is to place the three tasks of spatiotemporal function estimation, sensor localization, and sensor control under the common methodological umbrella of statistical data processing. The overall problem will be formulated by a statistical model that characterizes the statistical relations between spatiotemporal function, sensor positions, sensor measurements, and sensor control variables. Unknown or intractable aspects of the statistical model will be learned using deep neural networks. Based on this statistical model, advanced statistical data processing methods will be used to perform the three tasks of spatiotemporal function estimation, sensor localization, and sensor control in a consistent manner. This approach enables an exchange of probabilistic information between the three tasks, which allows the tasks to support each other and attain an unprecedented level of performance. Furthermore, our approach provides a quantitative characterization of the accuracy of spatiotemporal function estimation and sensor localization, which is needed in many applications such as mobile communications. The methods developed in the project will be implemented and tested for the real-world use case of estimating position- and time-dependent performance indicators in mobile communication networks. This use case is practically important because knowledge of performance indicators is required for the design and optimization of mobile communication networks in order to achieve reliable coverage levels and maintain high quality of service with limited resources. We expect that the results of the project will improve the accuracy of spatiotemporal function estimation and sensor localization while at the same time reducing the number and the power consumption of the sensors. This will push the performance boundaries of mobile sensor networks, especially in critical scenarios with severe power and range constraints. In particular, our results can be expected to improve the performance of future mobile communication networks.