Wireless Channel Prediction with Location Uncertainty
Spatial wireless channel prediction is important for future wireless networks, and
in particular for anticipatory networks to perform proactive resource allocation
at different layers of the protocol stack. In this thesis, we study location-aware
channel prediction with uncertainty in location information and understand its utilization to enhance the communication capabilities in wireless networks.
Paper A discusses challenges of 5G networks, which include an increase in traffic and number of devices, robustness for mission-critical services, and a reduction in total energy consumption and latency. We then argue how location information can be leveraged in addressing several of the key challenges in 5G with location-aware channel prediction by maintaining a channel database. We use Gaussian processes (GP) from machine learning in developing a framework for location-aware channel prediction. We then give a broad overview of using location-aware channel prediction in addressing the aforementioned challenges across different layers of the protocol stack.
In Paper B, we investigate two frameworks, classical Gaussian processes (cGP) and uncertain Gaussian processes (uGP), and analyze the impact of location uncertainty
during both training and testing. We have demonstrated that, when heterogeneous location uncertainties are present, the cGP framework is unable to (i) learn the underlying channel parameters properly; (ii) predict the expected channel quality metric. By introducing a GP that operates directly on the location distribution, we find uGP, which is able to both learn and predict in the presence of location uncertainties.
Paper C studies the tradeoffs in utilizing location information in the robust link scheduling problem (RLSP) at the medium access control layer. We compare two approaches to RLSP, one using channel gain estimates and the other using location information. Our comparison reveals that both approaches yield similar
performances, but with different overhead.