Browsing by Author "Hammarberg, Toni"

In this thesis we consider the problem of estimating the trajectory of a WiFi access point (AP), from nearby received signal strength measurements. This work is mainly motivated by the needs of WiFi-based localization, as WiFi-based localization is based on the presence of static (unmoving) APs, so detecting these moving APs and filtering them out is important. However, this work has some parallels to other field of research as well. The key challenge in this thesis is the data quality. WiFi-based localization often makes use of crowdsourced data, where we have almost no control of the data acquisition, making the data often much more noisy. For this reason we employ methods from Bayesian modeling, Markov chain Monte Carlo (MCMC) techniques in particular. Not only does Bayesian statistic allow robust framework for statistical modeling, but MCMC enables us to use many diagnostics to validate the model fit. Sometimes data can be poor enough that we should not base decisions on it, and the with MCMC diagnostics we can differentiate the successful cases from the cases where there is high uncertainty. Using these methods we are able to estimate the AP trajectory. However, the MCMC methods are computationally expensive, and since this work is motivated by real-world applications, computationally lighter methods capable of real-time performance would be desirable. For this reason we also investigate the use of variational inference (VI), but unfortunately the results leave much to be desired. In this thesis we conducted simple experiments to obtain real data, and the MCMC and VI data processing was done with the probabilistic programming language Stan.