An adaptive approach to mobile sampling

Radu Calin Gatej

AbstractThe increase in the number of hardware sensors that smartphones come equipped with in the past several years has determined the emergence of a new area for research and application development, called mobile sensing. Mobile sensing applications make use of the different sensors that a smartphone packs, to determine as much as possible about the context of a smartphone user. Location can be determined from the GPS sensor, physical activity could be determined from the accelerometer, or the presence of other users, using Bluetooth. Along the multitude of other sensors built into a smartphone, information about the user and his daily habits can be analyzed in order to help improve, for example, the users health. By combining together data gathered this way from groups of people and even entire communities, behavioral patterns for specific groups of people or for humans in general can be studied.
The issue that mobile sensing applications face is the impact that they have on the battery of a smartphone. Due to sampling a high number of sensors at the same time, some of them more battery costly than others (Global Positioning System sensor), the time in which data can be collected while the user mobile is limited. To reduce battery depletion, mobile sensing applications usually make a compromise by reducing the temporal resolution of sensor sampling. This thesis proposes an improvement in both battery life and temporal resolution of the data by using adaptive sampling, a technique which conditions sensor sampling on the context of the user, rather then periodically sampling the sensors. The solution is designed, implemented and tested to work with the SensibleDTU project, which is a mobile sensing experiment that aims to study the mobility and personal interaction of students at the Technical University of Denmark.
The proposed solution intends to decrease the use of the GPS sensor for location estimation, by using a local database of WiFi access points that have been already mapped, as WiFi is a cheaper sensor than GPS battery-wise. GPS scanning would thus be conditioned by the presence of such access points. The mapping of WiFi access points is designed and implemented in this thesis by combining WiFi and location data gathered over four months from 160 students.
The adaptive sampling scheme proposed in this project also adds the mobility of the user as a condition for triggering the GPS, since location data is more useful if the user changes his position.
Finally, efficiency of the implemented adaptive sampling scheme is evaluated in terms of battery usage, temporal resolution and accuracy of data.
TypeMaster's thesis [Academic thesis]
PublisherTechnical University of Denmark, Department of Applied Mathematics and Computer Science / DTU Co
AddressMatematiktorvet, Building 303B, DK-2800 Kgs. Lyngby, Denmark,
NoteDTU supervisors: Sune Lehmann, Jakob Eg Larsen, DTU Compute
Electronic version(s)[pdf]
Publication link
BibTeX data [bibtex]
IMM Group(s)Intelligent Signal Processing

Back  ::  IMM Publications