Oregon State University

Can’t find an event? We’re busy migrating to a new event calendar. Try looking at the new calendar



Event Details

MS Final Examination – Pingan Zhu

Monday, September 23, 2013 10:00 AM - 12:00 PM

Revenue-Based Spectrum Management via Markov Decision Process
We consider the problem of spectrum management in cognitive networks that maximizes the revenue for a spectrum operator. Specially, we study the problem on how a spectrum operator can optimally allocate its limited spectrum resource by arranging various classes users/devices who pays differently for per unit spectrum per unit time. We show that the problem of maximizing the revenue for the spectrum operator can be cast in the Markov Decision Process (MDP) framework. To that end, we investigate three formulations of MDP: the finite-horizon, discounted infinite-horizon models and monotone policy models. We show that for small scenarios, it is feasible to obtain the optimal solution for the finite horizon MDP using the backward induction algorithm. For larger scenarios, Q-learning is used to approximate the optimal solution for the discounted infinite horizon model. We also show an approach to compactly represent and predict optimal policies using support vector machine (SVM). Finally, we present that, if we can make certain assumptions on the system, we are able to give a threshold policy, which is not only optimal but also very efficient to implement. Simulation results are used to verify our approach.

Co-Major Advisor: Thinh Nguyen
Co-Major Advisor: Mina Ossiander
Committee: Alan Fern
Committee: Glencora Borradaile
GCR: Yevgeniy Kovchegov 

Kelley Engineering Center (campus map)
Nicole Thompson
1 541 737 7234
Nicole.Thompson at oregonstate.edu
Sch Elect Engr/Comp Sci
This event appears on the following calendars: