Opportunistic traffic Offloadings Mechanisms for Mobile/4G Networks

In the last few years, it has been observed a drastic surge of data traffic demand from mobile personal devices (smartphones and tablets) over cellular networks [1]. Even though a significant improvement in cellular bandwidth provisioning is expected with LTE-Advanced systems, the overall situation is not expected to change significantly. In fact, the diffusion of M2M and IoT devices is expected to increase at an exponential pace (the share of M2M devices is predicted to increase 5x by 2018 [1]) while the capacity of the cellular network is expected to increase linearly [1]. In order to meet such a high demand and to increase the capacity of the channel, multiple offloading techniques are currently under investigation, from modifications inside the cellular network architecture, to integration of multiple wireless broadband infrastructures, to exploiting direct communications between mobile devices. All these approaches can be diveded in two main classes: - To develop more sophisticated physical layer technologies (e.g. massive MIMO, higher-order modulation schemes, cooperative multi-period transmission/reception) - To offload part of the traffic from the cellular to another complementary network. From this perspective the thesis contributes on both areas. On the one hand we discuss our investigations about the performance of the LTE channel capacity through the development of a unified modelling framework of the MAC-level downlink throughput of a sigle LTE cell, which caters for wideband CQI feedback schemes, AMC and HARQ protocols as defined in the LTE standard. Furthemore we also propose a solution, based on reinforcement learning, to improve the LTE Adaptive Modulation and coding Scheme (MCS). On the other hand we have proposed and validated offloading mechanisms which are minimally invasive for users' mobile devices, as they use only minimally their resources. Furthemore, as opposed to most of the literature, we consider the case where requests for content are non-synchronised, i.e. users request content at random points in time.