Narrowband Radio Frequency Inference: Physical Modeling and Measurement Processing

Abstract

Narrowband RF inference is an emerging field that aims at estimating the location and actions of a person using commercial off-the-shelf narrowband wireless communication devices. In the research field, the location and actions of a person are estimated using the received signal strength measurements of the receivers. In this thesis, both statistical and deterministic measurement models are derived that can be used for processing the data to achieve the goals set by the applications. In particular, device-free localization and respiration rate monitoring are considered, and the models are used and extended for various scenarios. The results suggest that the models allow optimized deployments and low-complexity systems that can reach the performance of state-of-the-art systems.

The research in the thesis shows that the impact of a person on the measurements of a link can be represented using mutually exclusive three temporal states. This temporal characterization extends the widely used two-state model by adding a state to explicitly describe the reflection dominated the effect. The reflection model enables different possibilities, which are significant in a number practical aspects. First, the measurement of the links can be related to a large area around the link-line. This relation can be used for developing localization systems using as few as two receivers and can be used for detecting the occupancy of a region using a single receiver. Second, the impact of reflection can be parametrized using a single parameter known as excess path length. This parameter allows one to determine the effective area of the links and relate them to the measurements, which in turn enables development of a localization system that is detector based and has a low complexity. Third, the excess path length parametrization also enables one to investigate the impact of small amplitude movements, such as inhaling and exhaling induced chest movements. These implications are elaborated in depth and their validity is evaluated through experimentation.

The models presented in the thesis are elaborated numerically and validated using empirical measurement data. For the purpose, two different test subjects are used: a mobile robot equipped with a container that is simulating human torso, and a human. The former can navigate while accurately localizing itself in indoor environments, and has a repeatable set of physical parameters affecting the measurement. The latter allows one to observe the performance under realistic operating conditions. The effect of the test subjects on the received signal strength is measured using an experimental setup that is designed to acquire high-quality measurements. The acquired data are used for making conclusive statements about the validity of the models and assess the generality of the underlying assumptions. Therefore, the thesis is composed of both mathematical modeling and the development of the validation system.