Master Thesis: Retrospective Mobile Phone Position Estimation Using Crowdsourced Data Detail Records
Retrospective Mobile Phone Position Estimation Using
Crowdsourced Data Detail Records
Mobile phone networks consist of individual cells that are assigned
unique identifiers. Whenever a phone accesses the internet, exchanges
short messages, is used for a call or transmits signaling data, it
connects to the base station of such a network cell. Mobile phone
network operators log all these network events, including the
identifiers of the corresponding cells, both for billing purposes and
to assess the utilization of the network infrastructure. Since the
network operators know the geographical locations of the cells' base
stations, they can roughly estimate a phone's position for each event.
However, these position estimates are rather inaccurate due to the
large coverage areas of the network cells. Being able to more
accurately estimate the positions of the served phones would be
beneficial for the network operators. The main reason is that in many
countries, network operators are required by law to provide position
estimates for any mobile phone that is used to call an emergency
telephone number, because this location data helps in quickly
deploying the needed help, especially if the caller is not able to
communicate his or her location.
There are several different approaches to obtain more accurate position estimates of the served phones. Applying post-processing techniques to the introduced inaccurate estimates in order to improve their accuracy is an approach that is especially suitable for network operators. Therefore, different kinds of such post-processing techniques are evaluated in this thesis to answer the question how well suited the individual techniques are for the described task.
The thesis describes the basic working principles of mobile phone networks and provides the theoretical background for the estimation of mobile phone positions. On this basis, four post-processing techniques are proposed, each of them building upon a different fundamental idea. For the first technique, cell coverage areas are modeled by gaussian distributions. The second technique incorporates the land-use in the coverage area of the relevant network cell. In the third post-processing technique, a Kalman filter is applied to the available inaccurate position estimates. And lastly, the fourth technique combines the land-use based technique with the Kalman filtering technique.
After the introduction of the proposed techniques, the implementation of these techniques is discussed and their results are evaluated. The evaluation shows that all four techniques produce reasonably accurate position estimates. None of these techniques is completely preferable over the others, but
instead, each of the techniques is characterized by different strengths and weaknesses, which are discussed as part of the evaluation.
Supervisor: Friedhelm Victor 
Type: Master Thesis
Duration: 4 months
10587 Berlin, Germany
Phone: +49 30 8353 58811
Fax: +49 30 8353 58409