| dc.description.abstract | We are living in an era full of data services, and the advancement in statistical learning
encourages the development of intelligent system design algorithms based on practical
data. In our work, we plan to study two potential applications with intelligent design in
wireless systems based on statistical and machine learning techniques.
The first application we study is the spectrum sensing problem in energy harvesting
based cognitive radio networks, which is a promising solution to address the shortage of
both spectrum and energy. Since the spectrum access and power consumption pattern
are interdependent, and the power value harvested from certain environmental sources are
spatially correlated, the new power dimension could provide additional information to enhance
the spectrum sensing accuracy. In our work, the Markovian behavior of the primary
users is considered, based on which we adopt a hidden input Markov model to specify the
primary vs. secondary dynamics in the system. Accordingly, we propose a 2-D spectrum
vs. power (harvested) sensing scheme to improve the primary user detection performance,
which is also capable of estimating the primary transmit power level. Theoretical and
simulated results demonstrate the effectiveness of the proposed scheme, in terms of the
performance gain achieved by considering the new power dimension. To the best of our
knowledge, this is the first work to jointly consider the spectrum and power dimensions
for the cognitive primary user detection problem.
The second work is about spatio-temporal base station traffic prediction with machine
learning. Accurate prediction of user traffic in cellular networks is crucial to improve
the system performance in terms of energy efficiency and resource utilization. However,
existing work mainly considers the temporal traffic correlations within each cell while
neglecting the spatial correlation across neighboring cells. In this work, machine learning models that jointly explore the spatio-temporal correlations are proposed, where a multitask
learning approach is adopted to explore the commonalities and differences across cells
in improving the prediction performance. Base on real data, we demonstrate the benefits
of joint learning over spatial and temporal dimensions. | en |
