Stochastic optimization in OFDMA/LTE networks

Cellular networks are facing a considerable increase in mobile data traffic. To satisfy these demands, ambitious requirements are set for the current and upcoming generations of cellular networks. Frequency reuse of one is applied for a flexible usage of the resource allocation especially for asymmetrical load distributions. However, it limits the spectral efficiency of cell-edge located mobile terminals. In order to mitigate the interference and improve the capacity of cell-edge terminals, inter-cell interference coordination is used. This thesis considers the design of semi-static inter-cell interference coordination schemes for OFDMA/LTE networks. In this approach, base stations coordinate the power settings per resource block over long time spans such as seconds or even minutes. Due to the long time coordination period and the random evolution of fast fading and scheduling decisions, the overall system performance for a given power allocation becomes a random variable itself, turning the interference coordination problem into a stochastic optimisation problem where the expected system performance is to be maximised. Hence, we address two fundamental problems: (i) Modeling the expected system performance and (ii) Optimising the power allocations accordingly. We show that semi-static ICIC based on performance expectation models can be solved near-optimally and executed in real-time by means of an implementation in a graphic card. We deliver not only a system design but also mathematical expressions that describe the performance of interference limited systems operating on dynamic schedulers. Even for highly dynamic mobile scenarios, we can compute expectation values in real-time, as well as the associated power allocation.