Load-driven dynamic user assignment algorithms for dense cellular networks

A key option to further increase mobile network capacity is to deploy dense cellular networks (DCNs). This densification of cellular networks raises challenging issues though, as it likely increases the spatial variation and temporal fluctuations in load. To harness the full potential of DCNs, cell selection algorithms must take these varying load conditions into account. In this paper we study the optimal user association in DCNs based on a Linear Program (LP). Since several system parameters tend to be unknown and time-varying in practice, we develop a dynamic, self-organizing, and load-aware cell selection algorithm: the Shadow Price Assignment (SPA) algorithm. Our algorithm realizes an optimal user association without explicit knowledge of the system parameters by using a parsimonious set of dynamically adapted control parameters. We establish convergence of the control parameters under suitable assumptions. For larger systems the convergence may be slower, and we propose a local clustering approach to further improve the user-perceived performance in systems with many APs. Extensive simulations confirm that the SPA algorithm substantially outperforms conventional approaches.