Dissertation in the field of Signal processing for communications, Jan Oksanen

The title of thesis is ”Machine learning methods for spectrum exploration and exploitation”.

Practically all of our wireless communication takes place over radio frequency spectrum. In order for different wireless systems not to interfere with each other, they have been traditionally licensed for different frequency bands. The explosive growth of wireless traffic has, however, highlighted the weakness of such rigid frequency allocation. There are more and more wireless systems sharing the limited radio resources, and hence there will not be enough useful radio spectrum for all. This has spurred research on more flexible methods for utilizing the valua-ble radio spectrum, one of them being the concept of cognitive radio. Cognitive radio is a wireless device/system that allows for secondary use of the licensed spectrum. It identifies idle frequencies and accesses it in an opportunistic manner. Underutilized spectrum is a time-frequency-space varying resources.

In this thesis machine learning methods for identifying and exploiting idle spec-trum have been proposed and analyzed for cognitive radio. The proposed meth-ods are based on reinforcement learning and a problem formulation, where the frequency bands are seen as slot machines producing random profits. The pro-posed methods are easy to implement and provide an excellent trade-off be-tween simplicity and performance such as achieved data rate. Furthermore, methods based on reinforcement learning that allow for more efficient spectrum sensing by multiple cognitive users collaboratively are proposed. In the thesis the gains from collaborative spectrum sensing by multiple users for identifying idle spectrum have also been analyzed. These gains stem from spatial diversity by observing the same signals through multiple independent channels. The methods proposed in this thesis as well as the analytical results derived are useful for de-signing and analyzing future wireless communication networks ¬- particularly those that depend on spectral coexistence with other wireless systems.

Opponents: Professor Rick S. Blum, Lehigh University, USA and Professor Jean-Marie Gorce, National Institute of Applied Sciences (INSA) Lyon, France

Supervisor: Professor Visa Koivunen, Aalto University School of Electrical Engineering, Department of Signal Processing and Acoustics

Contact information
Jan Oksanen
+358 50 5408 889
jan.oksanen@aalto.fi